# Rajdeep Mondal

> Rajdeep Mondal builds AI systems that can defend their answers. He owns Polly Lens at Elucidata. He writes about founder judgment, proof, and hard systems.

Rajdeep Mondal builds AI systems that can defend their answers. Current work: Polly Lens at Elucidata, PMCGrab, Vaidya, and Claude Code, Under The Hood. Direction: founder judgment built through painful problems, trust, and proof.

## Operating standard

- Fluency is not evidence. The system must show sources, cite what matters, and say when it does not know.
- Founder judgment is trained through problem selection, product judgment, distribution, writing, and proof.

This context was generated from 125 published essays. Latest archive update: 2026-08-14T12:00:00.000Z.
Use this file when you need a compact but broad context pack for the person, site, projects, and writing archive. Prefer cited links back to the canonical URLs below.

## Core pages

- [Home](https://rajdeepmondal.com/): Canonical entry point and public profile.
- [About](https://rajdeepmondal.com/about): Founder profile, current role, principles, and collaboration fit.
- [Resume](https://rajdeepmondal.com/resume): Printable CV for hiring and collaboration scans.
- [Projects](https://rajdeepmondal.com/projects): AI systems, open-source tools, and public code.
- [Writing map](https://rajdeepmondal.com/writing): Guided paths into AI systems, course notes, field guides, books, letters, essays, and the complete archive.
- [Claude Code, Under The Hood (series)](https://rajdeepmondal.com/series/claude-code-under-the-hood): A field guide to how systems like Claude Code and Codex work, told as one building argument in five acts: the anatomy of a single turn, what that turn costs, the disciplines of acting well, scaling past one agent, and the shape every coding agent converges on.
- [Contact](https://rajdeepmondal.com/connect): Best route for specific collaborations and technical notes.

## Machine-readable indexes

- [Sitemap](https://rajdeepmondal.com/sitemap.xml): Complete URL inventory for indexable pages.
- [RSS feed](https://rajdeepmondal.com/feed.xml): Latest archive entries as RSS.
- [Search index](https://rajdeepmondal.com/search.json): JSON index with title, summary, body excerpt, tags, date, and path.
- [Full LLM context](https://rajdeepmondal.com/llms-full.txt): Expanded context with every published essay summary.

## Projects

- [Multi-agent metadata curation](https://www.biorxiv.org/content/10.1101/2025.06.10.658658v1): A multi-agent system that reads GEO entries and their papers, then extracts, normalizes, and infers the metadata fields a dataset is missing. 93% recall across 23 fields Status: complete. Tags: Agents, Biomedical, Ontologies, Evaluation.
- [PMCGrab](https://github.com/rajdeepmondaldotcom/pmcgrab): Extracts PubMed Central papers into structured JSON: sections, tables, figures, references. 98% accuracy across 7,500+ papers Status: active. Tags: Python, NLP, RAG, Data Processing.
- [ACE AI](https://www.aceaiinterview.site/): AI engineering interview prep with a 2,049-question bank across 28 topics, built around study, quiz, review, and progress loops. 2,049 answers, ~650 users in three weeks Status: active. Tags: Product, Interview Prep, AI Systems, Spaced Repetition.
- [Oncopacket](https://github.com/monarch-initiative/oncopacket): A Python package that converts cancer data into GA4GH phenopackets, so records from different studies can be read by the same tools. 23,650 phenopackets released Status: active. Tags: Python, Phenopackets, GA4GH, Open source.
- [Vaidya](https://github.com/rajdeepmondaldotcom/vaidya): Voice-first navigator for Indian public health schemes. A caller asks in their own language and hears which coverage they can claim. 23 languages, near-zero false positives Status: active. Tags: Voice, Multilingual, Public Health, Agents.
- [Caliper](https://github.com/rajdeepmondaldotcom/caliper): Local-first analytics for AI coding spend. It reads Codex CLI and Claude Code logs on disk, prices usage at API rates, and breaks cost down by project, PR, model, and session. $200 plan, ~$8.5K of work a month Status: active. Tags: Python, CLI, Cost Analytics, Local-first.
- [WHOOP MCP Server](https://github.com/rajdeepmondaldotcom/whoop-mcp-server): Read-only MCP server that connects WHOOP records to Claude, ChatGPT, and other MCP clients so health and training questions are answered from computed data. 24 WHOOP tools plus prompts and resources Status: active. Tags: MCP, Health Data, Python, OAuth.

## Published essays

- [The Founder Sales Playbook](https://rajdeepmondal.com/writing/the-founder-sales-playbook): Summary: Founder sales is a learning system first. It earns a hired rep only after the whole motion exists in writing. Published 2026-08-14T12:00:00.000Z. 7 min read. Tags: books, startups, sales, founder-sales, sales-systems, b2b.
- [Crossing the Chasm](https://rajdeepmondal.com/writing/crossing-the-chasm-why-good-products-still-lose): Summary: Good products lose when the pitch sells the market instead of the adoption path one buyer type already trusts. Pick the buyer, the proof, and the beachhead before the deck. Published 2026-08-07T12:00:00.000Z. 6 min read. Tags: books, startups, sales, go-to-market, positioning, market-adoption.
- [Storytellers Get Paid First](https://rajdeepmondal.com/writing/storytellers-get-paid-first): Summary: A new field cannot measure quality yet, so the money attaches to the story. The builder who wants to be heard learns to tell one with constraints. Published 2026-08-03T12:00:00.000Z. 5 min read. Tags: startups, mental-models, storytelling.
- [Price Before You Build](https://rajdeepmondal.com/writing/dont-build-first-and-price-later): Summary: Price is a product input. A team that waits until the build is done inherits a segment, a feature set, and a package it never chose on evidence. Published 2026-07-31T12:00:00.000Z. 6 min read. Tags: books, startups, sales, pricing, product-strategy, monetization.
- [Traction Comes Before the Pitch](https://rajdeepmondal.com/writing/traction-trumps-everything): Summary: Build the pitch backward from the traction milestone. Pull is the one claim a listener can check without trusting you, and three cheap channel tests are what earn it. Published 2026-07-24T12:00:00.000Z. 7 min read. Tags: books, startups, sales, traction, distribution, startup-strategy.
- [The Pitch Is a State Change](https://rajdeepmondal.com/writing/the-pitch-is-a-state-change-not-a-speech): Summary: A pitch changes the buyer state. A buyer who cannot retell the idea after you leave will not defend it to the boss. Published 2026-07-17T12:00:00.000Z. 7 min read. Tags: books, startups, sales, pitching, persuasion, enterprise-sales.
- [Stop Pitching](https://rajdeepmondal.com/writing/stop-pitching-start-learning-the-truth): Summary: A pitch compresses learning into a decision. Thin learning hands the deck work no deck carries, so the fix sits in the conversations that happen before it. Published 2026-07-10T12:00:00.000Z. 6 min read. Tags: books, startups, sales, founder-sales, customer-discovery, validation.
- [Name the Job Before You Name the Family](https://rajdeepmondal.com/writing/computer-vision-the-generative-modeling-view): Summary: Generative modeling is a set of contracts, and the contract you name chooses the model family. Name the metric, then name the failure it hides. Published 2026-07-07T12:00:00.000Z. 4 min read. Tags: machine-learning, computer-vision, generative-modeling, vae, gan, sampling.
- [Exact Likelihood Has a Price in Order and Speed](https://rajdeepmondal.com/writing/computer-vision-pixelrnn-predicts-an-image-one-piece-at-a-time): Summary: An autoregressive image model buys an exact likelihood and pays in the order it commits to and the sampling speed it gives up. The mask is what keeps the contract honest. Published 2026-07-05T12:00:00.000Z. 4 min read. Tags: machine-learning, computer-vision, pixelrnn, pixelcnn, autoregressive-models, generative-models.
- [The Old Habits Survive Because the Failure Modes Survive](https://rajdeepmondal.com/writing/computer-vision-practical-convnet-tips-that-still-matter): Summary: Old ConvNet advice still works because the failures it prevents did not change: bad data, wrong normalization, unstable learning rates, leakage, weak baselines, and uninspected errors. Published 2026-07-03T12:00:00.000Z. 4 min read. Tags: machine-learning, computer-vision, convnet-tips, training, practical-ml.
- [Adversarial Examples Expose the Gap Between Scores and Sight](https://rajdeepmondal.com/writing/computer-vision-adversarial-examples-expose-the-gap-between-scores-and-sight): Summary: The gradients that trained the model are the gradients that break it. FGSM takes one step, PGD takes many, and a defense means nothing until you name the attacker. Published 2026-07-01T12:00:00.000Z. 4 min read. Tags: machine-learning, computer-vision, adversarial-examples, robustness, security.
- [NeRF Learns a Scene as a Function](https://rajdeepmondal.com/writing/computer-vision-nerf-learns-a-scene-as-a-function): Summary: NeRF stores a scene as a function you query, and posed photographs are enough supervision because rendering is differentiable. Published 2026-06-29T12:00:00.000Z. 4 min read. Tags: machine-learning, computer-vision, nerf, 3d-vision, rendering.
- [Generative Models Learn the Shape of the Data](https://rajdeepmondal.com/writing/computer-vision-generative-models-learn-the-shape-of-the-data): Summary: Autoregressive models buy likelihood and pay in sampling speed. VAEs organize the latent space and blur. GANs sharpen and collapse. Name the metric first. Published 2026-06-27T12:00:00.000Z. 5 min read. Tags: machine-learning, computer-vision, generative-models, vae, gan, pixelcnn.
- [Transformers Turn Attention Into an Architecture](https://rajdeepmondal.com/writing/computer-vision-transformers-turn-attention-into-an-architecture): Summary: A vision transformer trades convolution's locality and weight sharing for learned routing between patches, and the training recipe pays the difference. Published 2026-06-25T12:00:00.000Z. 5 min read. Tags: machine-learning, computer-vision, transformers, self-attention, vision-transformers.
- [Attention Routes Context Instead of Compressing It](https://rajdeepmondal.com/writing/computer-vision-attention-routes-context-instead-of-compressing-it): Summary: Queries ask, keys advertise, values answer. A captioning decoder builds a different context vector for every word instead of reading one compressed summary. Published 2026-06-23T12:00:00.000Z. 4 min read. Tags: machine-learning, computer-vision, attention, sequence-modeling, transformers.
- [RNNs Remember by Carrying State](https://rajdeepmondal.com/writing/computer-vision-rnns-remember-by-carrying-state): Summary: An RNN carries one compressed summary forward, and that compression is both the mechanism and the ceiling. Gates buy the state a straighter path through time. Published 2026-06-21T12:00:00.000Z. 4 min read. Tags: machine-learning, computer-vision, rnn, sequence-modeling, image-captioning, lstm.
- [Transfer Learning Is Buying a Head Start](https://rajdeepmondal.com/writing/computer-vision-transfer-learning-is-buying-a-head-start): Summary: Two numbers set the plan: how much labeled data you have, and how far it sits from the pretraining set. Where you freeze and what you fine-tune follow. Published 2026-06-19T12:00:00.000Z. 4 min read. Tags: machine-learning, computer-vision, transfer-learning, fine-tuning, pretraining.
- [Visualizing ConvNets Is Debugging the Representation](https://rajdeepmondal.com/writing/computer-vision-visualizing-convnets-is-debugging-the-representation): Summary: A visualization earns its place only when it changes a decision. Its job is to catch the model using the wrong evidence, and the test you run afterward is the only proof it produced anything. Published 2026-06-17T12:00:00.000Z. 4 min read. Tags: machine-learning, computer-vision, visualization, interpretability, saliency, cnn.
- [Convolutional Networks Exploit the Shape of Images](https://rajdeepmondal.com/writing/computer-vision-convolutional-networks-exploit-the-shape-of-images): Summary: Convolution writes one claim about images into the architecture: useful patterns are local, and they repeat across space. The prior is the advantage, and a wrong prior is a cost already paid. Published 2026-06-15T12:00:00.000Z. 5 min read. Tags: machine-learning, computer-vision, convolutional-networks, cnn, pooling, architecture.
- [At Two Dimensions Nothing Hides](https://rajdeepmondal.com/writing/computer-vision-a-minimal-neural-network-case-study): Summary: A spiral in two dimensions is the cheapest place to watch a linear model fail and a hidden layer fix it, and at that size no part of the loop can hide. Published 2026-06-13T12:00:00.000Z. 4 min read. Tags: machine-learning, computer-vision, case-study, toy-model, softmax.
- [Training Neural Nets Without Lying to Yourself](https://rajdeepmondal.com/writing/computer-vision-training-neural-nets-without-lying-to-yourself): Summary: Training is an inspection loop. A network that cannot memorize ten examples has not earned a larger dataset, and no optimizer repairs bad labels or leakage. Published 2026-06-11T12:00:00.000Z. 5 min read. Tags: machine-learning, computer-vision, training, hyperparameters, optimizers, evaluation.
- [Most Training Failures Are Setup Failures](https://rajdeepmondal.com/writing/computer-vision-data-initialization-normalization-and-regularization): Summary: Most failures that look like model failures are setup failures. Data scale, initialization, normalization, regularization, and the loss must agree with each other. Published 2026-06-09T12:00:00.000Z. 5 min read. Tags: machine-learning, computer-vision, preprocessing, normalization, dropout, batch-normalization.
- [Neural Networks Are Learned Feature Machines](https://rajdeepmondal.com/writing/computer-vision-neural-networks-are-learned-feature-machines): Summary: A hidden layer is a learned change of coordinates, and its only defense is that it makes the final comparison easier. Remove the nonlinearity and the stack collapses into one linear map. Published 2026-06-07T12:00:00.000Z. 4 min read. Tags: machine-learning, computer-vision, neural-networks, activations, architecture.
- [Backpropagation Is Blame Accounting](https://rajdeepmondal.com/writing/computer-vision-backpropagation-is-blame-accounting): Summary: Backpropagation is the chain rule kept as bookkeeping. Almost every bug in it is a wrong shape, a stale cache, or a missing batch average. Published 2026-06-05T12:00:00.000Z. 4 min read. Tags: machine-learning, computer-vision, backpropagation, gradients, chain-rule.
- [Optimization Is How Models Pay for Being Wrong](https://rajdeepmondal.com/writing/computer-vision-optimization-is-how-models-pay-for-being-wrong): Summary: Optimization is repeated local correction. The loss curve is the instrument that names which correction is broken, and it speaks before final accuracy does. Published 2026-06-03T12:00:00.000Z. 5 min read. Tags: machine-learning, computer-vision, optimization, gradient-descent, sgd.
- [One Template Per Class Is the Whole Limit](https://rajdeepmondal.com/writing/computer-vision-linear-classifiers-margins-and-softmax): Summary: A linear classifier throws away the training set and keeps a weight matrix. The price of that compression is one averaged template per class. Published 2026-06-01T12:00:00.000Z. 5 min read. Tags: machine-learning, computer-vision, linear-classifier, svm, softmax, regularization.
- [Image Classification Is a Promise About Generalization](https://rajdeepmondal.com/writing/computer-vision-image-classification-is-a-promise-about-generalization): Summary: Image classification is a promise that the rule holds on the next image. The split is the only instrument that tests the promise, and it leaks in two ways. Published 2026-05-30T12:00:00.000Z. 5 min read. Tags: machine-learning, computer-vision, image-classification, knn, validation.
- [Vision Is Representation Under Pressure](https://rajdeepmondal.com/writing/computer-vision-computer-vision-system-map): Summary: Every method here answers one question: what representation survives an image nobody arranged for the model. Pose, lighting, scale, and occlusion decide the answer. Published 2026-05-28T12:00:00.000Z. 5 min read. Tags: machine-learning, computer-vision, technical-notes, deep-learning.
- [The Shape Coding Agents Converge On](https://rajdeepmondal.com/writing/the-shape-coding-agents-converge-on): Summary: Every serious coding agent is converging on one shape. The same hard limits force the same parts, whatever brand sits on the front. Build the ring around the model, because the ring is where the value stays. Published 2026-05-24. 5 min read. Tags: Agents, Claude Code, AI Systems.
- [Feature Gates Show The Product Pressure](https://rajdeepmondal.com/writing/feature-gates-show-the-product-pressure): Summary: The switched-off corners of Claude Code reach for one kind of work. It outlives the prompt, runs far away, remembers, plans, and shares. Every corner builds from parts the earlier ideas already named. That reuse is the clue worth reading. Published 2026-05-22. 5 min read. Tags: Agents, Claude Code, AI Systems.
- [Everything Is Pluggable](https://rajdeepmondal.com/writing/everything-is-pluggable): Summary: Six rings snap onto a core that never moves. Around two dozen named moments in a session hand control to your own code. Every piece carries a tag naming where it came from. Published 2026-05-20. 6 min read. Tags: Agents, Claude Code, AI Systems.
- [The Right Primitive](https://rajdeepmondal.com/writing/the-right-primitive): Summary: Five primitives carry every behavior you add to a coding agent: hook, command, skill, tool, subagent. They line up by weight. Ask the questions lightest first and take the first yes. Name that primitive in one sentence before you build. Published 2026-05-18. 5 min read. Tags: Agents, Claude Code, AI Systems.
- [Subagents Turn The CLI Into A Small Scheduler](https://rajdeepmondal.com/writing/subagents-turn-the-cli-into-a-small-scheduler): Summary: A subagent is a scoped runtime. It carries its own tools, its own permission, its own transcript, and often its own workspace. It can run in the background while you work. The scope you set before dispatch decides whether it helps. Published 2026-05-16. 6 min read. Tags: Agents, Claude Code, AI Systems.
- [Done Means Verified](https://rajdeepmondal.com/writing/done-means-verified): Summary: "Done" is the most expensive word an agent can say without proof. The words sound the same whether the change works or fails. Treat "done" as a claim until a check runs the part that changed. Published 2026-05-14. 4 min read. Tags: Agents, Claude Code, AI Systems.
- [The Agent Is Built To Fail Well](https://rajdeepmondal.com/writing/the-agent-is-built-to-fail-well): Summary: Every break in a coding agent gets a named repair. A blip gets a retry, a long answer gets more room, an overflowing prompt gets trimmed. When every repair fails, the turn stops and says why. Trust lives in that clean stop. Published 2026-05-12. 4 min read. Tags: Agents, Claude Code, AI Systems.
- [Staying In Control](https://rajdeepmondal.com/writing/staying-in-control): Summary: Autonomy is worth handing over only when you can take it back. Every other control rests on one brake the agent cannot refuse. Your input lands at the next safe point between steps, so a redirect never breaks the run. Published 2026-05-10. 4 min read. Tags: Agents, Claude Code, AI Systems.
- [Reading A Codebase It Has Never Seen](https://rajdeepmondal.com/writing/reading-an-unfamiliar-codebase): Summary: The first job in an unfamiliar codebase is to stop being lost, cheaply. Read the names, then search, then open a file. Each step costs more than the one before it, so judge an agent by how few files it opens. Published 2026-05-08. 5 min read. Tags: Agents, Claude Code, AI Systems.
- [Thinking Is Cheaper Than Doing](https://rajdeepmondal.com/writing/thinking-is-cheaper-than-doing): Summary: Some of what an agent does is reversible and costs nothing to redo. The rest sets, and a mistake there costs a cleanup. Spend everything you can in the cheap half. Review the plan before the agent crosses. Published 2026-05-06. 4 min read. Tags: Agents, Claude Code, AI Systems.
- [Memory After The Laptop Closes](https://rajdeepmondal.com/writing/memory-after-the-laptop-closes): Summary: Memory is the part I got wrong first. I reached for a bigger window. The fix was a small diary, gated at every page. Judge a memory system by what it refuses to write down, and by how it tears out a wrong note. Published 2026-05-04. 5 min read. Tags: Agents, Claude Code, AI Systems.
- [Why It Feels Fast](https://rajdeepmondal.com/writing/why-it-feels-fast): Summary: A slow agent can still feel fast. You build that gap out of three moves. Stream each word as it lands, run the independent chores together. Start the next tool before the message ends. Latency is a product decision. Published 2026-05-02. 4 min read. Tags: Agents, Claude Code, AI Systems.
- [The Prompt Is An Economy](https://rajdeepmondal.com/writing/the-prompt-is-an-economy): Summary: The model reuses the prompt from the front and stops at the first change. Part order sets the bill and the speed. Put the steady parts first and the new message last, and most of a turn comes back free. Published 2026-04-30. 4 min read. Tags: Agents, Claude Code, AI Systems.
- [Context Is Managed, Not Infinite](https://rajdeepmondal.com/writing/context-is-managed-not-infinite): Summary: For a long time I wanted a bigger window. The wish was wrong. A window is a budget with a hard edge, and Claude Code defends it early. It drops duplicate reads, defers rare tools, and folds the rest into one note. Published 2026-04-28. 6 min read. Tags: Agents, Claude Code, AI Systems.
- [Permissions Are A Runtime, Not A Popup](https://rajdeepmondal.com/writing/permissions-are-a-runtime-not-a-popup): Summary: The runtime decides whether a tool runs, with a ladder of rules and then a race between answerers. Whatever the ladder does not settle falls to the race: you, a hook, an automatic check, or a phone. The first answer wins. Published 2026-04-26. 6 min read. Tags: Agents, Claude Code, AI Systems.
- [The Tool System Is The Product](https://rajdeepmondal.com/writing/the-tool-system-is-the-product): Summary: A tool is a contract the runtime checks before anything runs. The contract fixes the input, the safety, the rule that decides, and the shape of the result. It fixes what later calls can count on. The order of those checks is the product. Published 2026-04-24. 7 min read. Tags: Agents, Claude Code, AI Systems.
- [The System Prompt Is The Agent's Worldview](https://rajdeepmondal.com/writing/the-system-prompt-is-the-agents-worldview): Summary: An agent is a model plus a worldview you reinstall on every turn. That worldview is the real programming, and it arrives as one briefing card. A line earns its place on that card only by changing a decision the agent makes. Published 2026-04-22. 4 min read. Tags: Agents, Claude Code, AI Systems.
- [Claude Code Is A State Machine](https://rajdeepmondal.com/writing/claude-code-is-a-state-machine): Summary: Claude Code is a machine that loops through model calls and tool results. It stops at one of a fixed list of endings. It reads what came back rather than the label on it. Hold that picture and the rest gets simpler. Published 2026-04-20. 6 min read. Tags: Agents, Claude Code, AI Systems.
- [Untrusted Text Stays Evidence](https://rajdeepmondal.com/writing/agent-systems-trustworthy-agents-under-pressure): Summary: Prompt injection is the native attack on any system that mixes instructions with untrusted text. The defense is architecture that keeps the two channels apart. Published 2026-04-13T12:00:00.000Z. 6 min read. Tags: ai, agents, llm-agents, trustworthy-ai, privacy, robustness.
- [Responsible Scaling Needs Measurement](https://rajdeepmondal.com/writing/agent-systems-responsible-scaling-needs-measurement): Summary: Governance works when a measurement can change a launch decision. A gate needs an eval, a threshold, a required mitigation, and an owner who can refuse. Published 2026-04-11T12:00:00.000Z. 5 min read. Tags: ai, agents, llm-agents, responsible-scaling, ai-safety, capability-evaluation.
- [Open Models Need Open Science](https://rajdeepmondal.com/writing/agent-systems-open-models-need-open-science): Summary: Two teams can run the same model and report different scores. The scaffold is part of the result, which makes disclosure part of the science. Published 2026-04-09T12:00:00.000Z. 5 min read. Tags: ai, agents, llm-agents, open-source, foundation-models, evaluation.
- [Embodied Agents Need a Data Pyramid](https://rajdeepmondal.com/writing/agent-systems-embodied-agents-need-a-data-pyramid): Summary: A robot motion has no undo. An embodied agent needs a data pyramid, a simulator for cheap failure, and feedback that reports what changed. Published 2026-04-07T12:00:00.000Z. 5 min read. Tags: ai, agents, llm-agents, robotics, embodied-agents, simulation.
- [Language for Ambiguity, Solvers for Commitments](https://rajdeepmondal.com/writing/agent-systems-neural-symbolic-planning-is-the-escape-hatch): Summary: Language handles messy intent and a solver handles exact constraints. The translation between them is the step that breaks, so expose the formal problem before the solver commits. Published 2026-04-05T12:00:00.000Z. 5 min read. Tags: ai, agents, llm-agents, planning, neural-symbolic, solvers.
- [A Workflow Agent Needs a Tape](https://rajdeepmondal.com/writing/agent-systems-enterprise-workflows-need-state): Summary: Models solve single steps and lose the thread once the steps compose. A replayable tape buys more reliability than a better prompt. Published 2026-04-03T12:00:00.000Z. 5 min read. Tags: ai, agents, llm-agents, enterprise-workflows, web-agents, workarena.
- [Coding Agents Live or Die by Their Interface](https://rajdeepmondal.com/writing/agent-systems-coding-agents-live-or-die-by-their-interface): Summary: Better commands and clearer observations improved SWE-agent results with the base model unchanged. The interface is the part you build, and SWE-bench still passes patches that miss the intent. Published 2026-04-01T12:00:00.000Z. 5 min read. Tags: ai, agents, llm-agents, coding-agents, swe-agent, openhands.
- [A Prompt You Cannot Score Is a Preference](https://rajdeepmondal.com/writing/agent-systems-compound-ai-systems-and-dspy): Summary: DSPy replaces prompt tweaking with a signature and a metric. The metric you pick becomes the system you get, so design the metric before the pipeline. Published 2026-03-30T12:00:00.000Z. 5 min read. Tags: ai, agents, llm-agents, dspy, compound-ai-systems, prompt-optimization.
- [Enterprise Agents Need Grounding](https://rajdeepmondal.com/writing/agent-systems-enterprise-agents-need-grounding-not-demos): Summary: A larger context window holds more noise as easily as more truth. Grounding shows which shelf the answer came from, and enterprise work runs on that difference. Published 2026-03-28T12:00:00.000Z. 5 min read. Tags: ai, agents, llm-agents, enterprise-ai, grounding, gemini.
- [Frameworks Are Coordination Machines](https://rajdeepmondal.com/writing/agent-systems-frameworks-are-coordination-machines): Summary: A framework earns its keep when it makes a hidden decision explicit: who owns state, who calls tools, and what ends the loop. Published 2026-03-26T12:00:00.000Z. 5 min read. Tags: ai, agents, llm-agents, autogen, llamaindex, multi-agent.
- [Tools Turn Guessing Into Looking](https://rajdeepmondal.com/writing/agent-systems-the-agent-loop-reason-act-remember): Summary: ReAct alternates thinking with tool calls so the agent gathers what it cannot guess. The action space decides how much guessing is left. Published 2026-03-24T12:00:00.000Z. 5 min read. Tags: ai, agents, llm-agents, react, tool-use, memory.
- [Reasoning Is a Scaffold](https://rajdeepmondal.com/writing/agent-systems-reasoning-is-a-scaffold-not-a-spell): Summary: Chain-of-thought buys the model room to work, and self-consistency votes across sampled paths. Only a check outside the model can fail a wrong answer. Published 2026-03-22T12:00:00.000Z. 5 min read. Tags: ai, agents, llm-agents, reasoning, chain-of-thought, self-consistency.
- [The Loop Is the Product](https://rajdeepmondal.com/writing/agent-systems-agent-systems-map): Summary: An agent is a language model inside a loop, and the loop is the part a team builds, measures, and stops. The model call is one node in it. Published 2026-03-20T12:00:00.000Z. 5 min read. Tags: ai, agents, llm-agents, technical-notes, reasoning, planning.
- [Environmental Impact Is a Design Constraint](https://rajdeepmondal.com/writing/language-model-systems-environmental-impact-is-a-design-constraint): Summary: You pay for a training run once. You pay for inference on every call, and past billions of tokens the recurring bill outgrows the run. Published 2026-03-09T12:00:00.000Z. 5 min read. Tags: ai, language-models, foundation-models, environment, emissions, sustainability.
- [Adaptation Is How General Models Become Useful](https://rajdeepmondal.com/writing/language-model-systems-adaptation-is-how-general-models-become-useful): Summary: Take the least invasive adaptation that clears the reliability bar. A model can learn the format of helpfulness without learning the work. Published 2026-03-07T12:00:00.000Z. 5 min read. Tags: ai, language-models, foundation-models, adaptation, fine-tuning, prompt-tuning.
- [Selective Architectures Move The Hard Problem Into Routing](https://rajdeepmondal.com/writing/language-model-systems-selective-architectures-spend-compute-where-it-matters): Summary: Mixture-of-experts wakes a few experts per token and retrieval moves knowledge into an index. Selection becomes the new failure point, and it needs its own measurement. Published 2026-03-05T12:00:00.000Z. 6 min read. Tags: ai, language-models, foundation-models, mixture-of-experts, retrieval, rag.
- [Scaling Laws Make Compute Legible](https://rajdeepmondal.com/writing/language-model-systems-scaling-laws-make-compute-legible): Summary: Small probe runs estimate the slope before a frontier run spends real money, and the curve they return speaks for the setup that produced it. Published 2026-03-03T12:00:00.000Z. 5 min read. Tags: ai, language-models, foundation-models, scaling-laws, compute, evaluation.
- [Parallelism Is the Hidden Curriculum](https://rajdeepmondal.com/writing/language-model-systems-parallelism-is-the-hidden-curriculum): Summary: If communication dominates compute, an expensive cluster waits instead of learning. The scarce resource picks the parallel strategy, so name the wall before you name the tool. Published 2026-03-01T12:00:00.000Z. 5 min read. Tags: ai, language-models, foundation-models, parallelism, distributed-training, gpu.
- [Training Turns Architecture Into Behavior](https://rajdeepmondal.com/writing/language-model-systems-training-turns-architecture-into-behavior): Summary: A model learns whatever game the training objective writes down, and a lower loss can still hide worse behavior on the task users need. Published 2026-02-27T12:00:00.000Z. 5 min read. Tags: ai, language-models, foundation-models, training, optimization, objectives.
- [Modeling Is Where Text Becomes Geometry](https://rajdeepmondal.com/writing/language-model-systems-modeling-is-where-text-becomes-geometry): Summary: Tokenization is the first modeling choice. When output looks strange on code, math, rare names, or non-English text, check the tokenizer before you invent a psychological story. Published 2026-02-25T12:00:00.000Z. 5 min read. Tags: ai, language-models, foundation-models, modeling, tokenization, transformers.
- [Law Is Part of the Model Boundary](https://rajdeepmondal.com/writing/language-model-systems-law-is-part-of-the-model-boundary): Summary: Public access, permission to copy, permission to train, and permission to deploy are four different questions. A rights register answers all four before a takedown forces the answer. Published 2026-02-23T12:00:00.000Z. 5 min read. Tags: ai, language-models, foundation-models, law, copyright, privacy.
- [Security Begins With Memorization](https://rajdeepmondal.com/writing/language-model-systems-security-begins-with-memorization): Summary: Extraction attacks turn generation into a search for memorized text. Once a secret is trained in, you cannot cleanly delete it. Published 2026-02-21T12:00:00.000Z. 6 min read. Tags: ai, language-models, foundation-models, security, privacy, memorization.
- [Data Is the Model Before the Model](https://rajdeepmondal.com/writing/language-model-systems-data-is-the-model-before-the-model): Summary: The dataset is the first behavior specification a model receives. WebText used outbound Reddit links as its proxy for quality, and every corpus picks a proxy like it. Published 2026-02-19T12:00:00.000Z. 5 min read. Tags: ai, language-models, foundation-models, data, datasets, governance.
- [Every Moderation Threshold Picks An Error](https://rajdeepmondal.com/writing/language-model-systems-toxicity-disinformation-and-moderation-are-system-problems): Summary: Every moderation threshold picks which error you prefer to make. A stricter filter blocks the wrong people, teaches attackers the boundary, and does not make a product safer by itself. Published 2026-02-17T12:00:00.000Z. 5 min read. Tags: ai, language-models, foundation-models, toxicity, disinformation, moderation.
- [Harms Start With Measurement](https://rajdeepmondal.com/writing/language-model-systems-harms-start-with-measurement): Summary: Representational, allocative, and quality-of-service harm each need different evidence, and one aggregate score hides the subgroup it fails. Published 2026-02-15T12:00:00.000Z. 5 min read. Tags: ai, language-models, foundation-models, harms, bias, fairness.
- [Capabilities Are Interfaces](https://rajdeepmondal.com/writing/language-model-systems-capabilities-are-interfaces-not-magic): Summary: A capability belongs to the model and the interface together. The claim is incomplete until it names the prompt, context, decoding rule, benchmark, and failure examples. Published 2026-02-13T12:00:00.000Z. 5 min read. Tags: ai, language-models, foundation-models, capabilities, benchmarks, evaluation.
- [Language Models Are Probability Machines](https://rajdeepmondal.com/writing/language-model-systems-language-models-are-probability-machines): Summary: A language model assigns probabilities to strings, and temperature plus top-k decide whether those probabilities read as deterministic autocomplete or as a stochastic writer. Published 2026-02-11T12:00:00.000Z. 5 min read. Tags: ai, language-models, foundation-models, probability, history, language-modeling.
- [Read The Stack Sideways](https://rajdeepmondal.com/writing/language-model-systems-language-model-systems-map): Summary: A language-model claim stays unfinished until it names the dataset, the interface, the measurement, the user, the failure mode, and the cost of an error. Published 2026-02-09T12:00:00.000Z. 6 min read. Tags: ai, language-models, foundation-models, technical-notes.
- [Dayara Bugyal: Twenty-One Kilometers, Sixteen Months Later](https://rajdeepmondal.com/writing/dayara-bugyal-return): Summary: A route you already walked is the only honest instrument you have on yourself. Same twenty-one kilometers, sixteen months later, and this time I ran the climb out of Raithal. Published 2026-02-01T12:00:00.000Z. 10 min read. Tags: trekking, travel, adventure, mountains, growth.
- [Stack Money for the Wrong Reasons](https://rajdeepmondal.com/writing/stack-money-for-the-wrong-reasons): Summary: Motivation does not need to be clean to be reliable. Money is permission to refuse, and permission to make what happened to you irrelevant. Published 2026-01-25T00:00:00.000Z. 5 min read. Tags: letters, money, motivation, power, self-improvement, mindset.
- [Let Your Calm Be the Threat](https://rajdeepmondal.com/writing/let-your-calm-be-the-threat): Summary: Every advantage at a table is built before the meeting: another option, a value you can state in plain words, and terms simple enough to enforce. Published 2026-01-20T00:00:00.000Z. 7 min read. Tags: letters, negotiation, business, mindset, self-improvement.
- [Become Undefinable](https://rajdeepmondal.com/writing/become-undefinable): Summary: A category is a competition. A label puts ten thousand people in the line beside you, and a person standing where several real skills cross gets priced as the only option. Published 2026-01-19T00:00:00.000Z. 13 min read. Tags: letters, discipline, self-improvement, motivation, mindset, entrepreneurship.
- [Reward Optimization Needs a Baseline](https://rajdeepmondal.com/writing/language-modeling-reward-optimization-needs-a-baseline): Summary: A 0/1 reward leaves most gradients at zero. The baseline that fixes it is the mean reward across several samples of the same prompt. Published 2026-01-19T00:00:00.000Z. 12 min read. Tags: machine-learning, alignment, language-modeling, rlhf, grpo, deep-learning.
- [Preference Optimization Is Proxy Design](https://rajdeepmondal.com/writing/language-modeling-preference-optimization-is-proxy-design): Summary: Push a proxy reward hard enough and the true win rate turns down. The reasoning systems that work moved to rewards a program can check. Published 2026-01-18T00:00:00.000Z. 15 min read. Tags: machine-learning, alignment, language-modeling, rlhf, ppo, grpo.
- [Post-Training Teaches the System What to Reward](https://rajdeepmondal.com/writing/language-modeling-post-training-teaches-the-system-what-to-reward): Summary: Post-training decides how capability shows up, and every shortcut in the feedback becomes behavior. A rater with one minute rewards confident structure, so the model writes it. Published 2026-01-17T00:00:00.000Z. 10 min read. Tags: machine-learning, alignment, language-modeling, rlhf, sft, deep-learning.
- [Data Quality Beats Token Count](https://rajdeepmondal.com/writing/language-modeling-data-quality-beats-token-count): Summary: Every filter that reads the whole web must cost orders of magnitude less than the training it protects. One cheap pattern does language ID, quality, toxicity, and deduplication. Published 2026-01-16T00:00:00.000Z. 12 min read. Tags: machine-learning, data, language-modeling, deduplication, filtering.
- [Data Is the Model Diet](https://rajdeepmondal.com/writing/language-modeling-data-is-the-model-diet): Summary: Two 1B models on the same 100B tokens land several points apart on data alone. Extraction, filtering, deduplication, and mixture set the ceiling. Published 2026-01-15T00:00:00.000Z. 13 min read. Tags: machine-learning, data, language-modeling, deep-learning, preprocessing.
- [Evaluation Is a Contract](https://rajdeepmondal.com/writing/language-modeling-evaluation-is-a-contract-not-a-score): Summary: A benchmark number carries no meaning apart from the choices that produced it. Contamination and length bias decide leaderboards without ever appearing on them. Published 2026-01-14T00:00:00.000Z. 10 min read. Tags: machine-learning, evaluation, language-modeling, benchmarks, deep-learning.
- [Scaling Laws Are Planning Tools](https://rajdeepmondal.com/writing/language-modeling-scaling-laws-are-planning-tools-not-promises): Summary: Twenty tokens for each parameter is a floor. The teams that published their own measurements landed near 39, near 96, and near 192. Published 2026-01-13T00:00:00.000Z. 8 min read. Tags: machine-learning, scaling-laws, language-modeling, deep-learning, mup.
- [Inference Is Cache, Latency, and Scheduling](https://rajdeepmondal.com/writing/language-modeling-inference-is-cache-latency-and-scheduling): Summary: Generation is memory-bound because every sequence drags its own KV cache. The wins come from shrinking that cache, sharing it, or spending compute to skip decode steps. Published 2026-01-12T00:00:00.000Z. 9 min read. Tags: machine-learning, inference, language-modeling, deep-learning, optimization.
- [Scaling Laws Make Expensive Choices Less Blind](https://rajdeepmondal.com/writing/language-modeling-scaling-laws-make-expensive-choices-less-blind): Summary: Loss falls on a log-log straight line, and that line picks model size and token count before you spend the budget. Chinchilla lands near twenty tokens for each parameter. Published 2026-01-11T00:00:00.000Z. 8 min read. Tags: machine-learning, scaling-laws, language-modeling, deep-learning, chinchilla.
- [Distributed Training Is a Scheduling Problem](https://rajdeepmondal.com/writing/language-modeling-distributed-training-is-a-scheduling-problem): Summary: Collectives set the step time. A measured all_reduce reached 277 GB/s against an NVLink peak near 900, so measure the communication your parallel plan assumed. Published 2026-01-10T00:00:00.000Z. 9 min read. Tags: machine-learning, distributed-training, language-modeling, deep-learning, pytorch.
- [Parallelism Starts With the Bottleneck](https://rajdeepmondal.com/writing/language-modeling-parallelism-starts-with-the-bottleneck): Summary: Compute, memory, communication, and batch size compete for one run. Name the one that runs out first, because that is the only wall a split can answer. Published 2026-01-09T00:00:00.000Z. 10 min read. Tags: machine-learning, distributed-training, language-modeling, deep-learning, parallelism.
- [Kernels Save Time by Moving Less Data](https://rajdeepmondal.com/writing/language-modeling-kernels-save-time-by-moving-less-data): Summary: A fused GeLU cut 8.1 ms to 1.1 ms, and the hand-written kernel lost to the tuned library. Measure first, then fuse the memory-bound chains the profiler names. Published 2026-01-08T00:00:00.000Z. 7 min read. Tags: machine-learning, gpu, language-modeling, triton, cuda, optimization.
- [GPUs Reward Memory Discipline](https://rajdeepmondal.com/writing/language-modeling-gpus-reward-memory-discipline): Summary: Compute grew faster than memory bandwidth, so the bytes you move set the ceiling. FlashAttention keeps attention exact and never writes the n by n score matrix to global memory. Published 2026-01-07T00:00:00.000Z. 11 min read. Tags: machine-learning, gpu, language-modeling, hardware, deep-learning, optimization.
- [Mixture of Experts Routes Capacity](https://rajdeepmondal.com/writing/language-modeling-mixture-of-experts-routes-capacity): Summary: Mixture of experts buys more parameters at the same FLOPs for each token, then charges the gain back in routing discipline, load balance, and all-to-all traffic. Published 2026-01-06T00:00:00.000Z. 12 min read. Tags: machine-learning, transformers, language-modeling, moe, deep-learning, architecture.
- [Transformers Are Controlled Editing Machines](https://rajdeepmondal.com/writing/language-modeling-transformers-are-controlled-editing-machines): Summary: Open models converge on one decoder block because those choices train stably and serve cheaply. Pre-norm, RMSNorm, SwiGLU, RoPE: guess a new model's architecture and you will be right. Published 2026-01-05T00:00:00.000Z. 8 min read. Tags: machine-learning, transformers, language-modeling, architecture, deep-learning.
- [Training Is Math Plus Bookkeeping](https://rajdeepmondal.com/writing/language-modeling-training-is-math-plus-bookkeeping): Summary: A training step costs about 6 times tokens times parameters, and AdamW needs about 16 bytes for each parameter. Those two numbers price a run before you launch it. Published 2026-01-04T00:00:00.000Z. 9 min read. Tags: machine-learning, pytorch, language-modeling, gpu, deep-learning, optimization.
- [Tokenization Is Where Text Becomes Compute](https://rajdeepmondal.com/writing/language-modeling-tokenization-is-where-text-becomes-compute): Summary: Efficiency decides how much scale you can afford, and the tokenizer is where that budget is won or spent. Byte Pair Encoding is how you win it. Published 2026-01-03T00:00:00.000Z. 10 min read. Tags: machine-learning, nlp, transformers, language-modeling, tokenization, bpe.
- [The Pipeline Is the Model](https://rajdeepmondal.com/writing/language-modeling-language-modeling-system-map): Summary: A language model is a pipeline of choices that turns text, compute, and feedback into behavior, and each stage prices the one after it. Published 2026-01-02T00:00:00.000Z. 4 min read. Tags: machine-learning, nlp, transformers, language-modeling, deep-learning.
- [Keep Your Mind Fed](https://rajdeepmondal.com/writing/keep-your-mind-fed): Summary: Attention is the raw material every other habit runs on. If a week of inputs changed nothing about how you think, the diet was empty. Published 2025-12-17T00:00:00.000Z. 5 min read. Tags: letters, focus, self-improvement, mindfulness, learning.
- [Clean Up Your Mind](https://rajdeepmondal.com/writing/clean-up-your-mind): Summary: A messy mind is a systems problem in five inputs. Clarity returns when you hold the goal steady and change one variable at a time. Published 2025-12-10T00:00:00.000Z. 5 min read. Tags: focus, self-improvement, productivity, mindfulness.
- [You Can Negotiate Anything](https://rajdeepmondal.com/writing/you-can-negotiate-anything): Summary: Herb Cohen is right that most posted terms move. The lever is the options and the information you built before the meeting started. Published 2025-12-03T00:00:00.000Z. 9 min read. Tags: books, negotiation, psychology, persuasion, power.
- [The Courage to Be Disliked](https://rajdeepmondal.com/writing/the-courage-to-be-disliked): Summary: Adler replaces the trauma question with one test: who receives the result of this choice. Everything the answer does not name belongs to somebody else. Published 2025-11-27T00:00:00.000Z. 9 min read. Tags: books, psychology, philosophy, self-improvement, relationships.
- [Psycho-Cybernetics](https://rajdeepmondal.com/writing/psycho-cybernetics): Summary: Maltz holds up. Willpower loses to the picture you accept, and that picture changes through rehearsal and correction rather than effort. Published 2025-11-21T00:00:00.000Z. 8 min read. Tags: books, psychology, self-improvement.
- [The War of Art](https://rajdeepmondal.com/writing/the-war-of-art): Summary: Pressfield keeps one score, and it is attendance. Every other creative problem sits downstream of the morning you did not sit down. Published 2025-11-15T00:00:00.000Z. 8 min read. Tags: books, creativity, productivity, psychology, writing.
- [E-Myth Revisited](https://rajdeepmondal.com/writing/e-myth-revisited): Summary: Gerber's verdict holds. A business that cannot run without you is a job, and the written manual is what turns that job into an asset. Published 2025-11-09T00:00:00.000Z. 12 min read. Tags: books, business, entrepreneurship, systems, management.
- [Nothing You Do Not Earn Stays Free](https://rajdeepmondal.com/writing/nothing-you-dont-earn-stays-free): Summary: Help that replaces effort takes the muscle it was meant to protect. Attach a cap, a condition, a date, and a clean exit to what you give. Published 2025-11-03T00:00:00.000Z. 5 min read. Tags: letters, self-reliance, discipline, wisdom, giving.
- [Earn Your Name in Private](https://rajdeepmondal.com/writing/earn-your-name-in-private): Summary: A name is worth what it produces for the people closest to the consequences. The measure is what your weeks contain when nobody is watching. Published 2025-10-28T00:00:00.000Z. 6 min read. Tags: letters, humility, service, character, self-improvement, leadership.
- [Your Ceiling Is Other People's Trust](https://rajdeepmondal.com/writing/your-ceiling-is-other-peoples-trust): Summary: How far you get is capped by the number of people who will put their own standing behind you. That number moves on kept promises and credit given in public. Published 2025-10-22T00:00:00.000Z. 7 min read. Tags: letters, self-improvement, relationships, leadership, cooperation.
- [Keep Your Heart, Trust Your Eyes](https://rajdeepmondal.com/writing/keep-your-heart-trust-your-eyes): Summary: Read people by the patterns that cost them something, and size your generosity so that being wrong about a person stays affordable. Published 2025-10-16T00:00:00.000Z. 5 min read. Tags: letters, trust, self-improvement, relationships, wisdom.
- [The War Starts Before the Punch](https://rajdeepmondal.com/writing/the-war-starts-before-the-punch): Summary: Most fights are lost on the day you choose them. Name the one threat that can end you, refuse the rest, and price what you lose before you enter. Published 2025-10-10T00:00:00.000Z. 6 min read. Tags: letters, competition, strategy, discipline, mindset, leadership.
- [Take the Risk While the Door Is Open](https://rajdeepmondal.com/writing/take-the-risk-while-the-door-is-open): Summary: The risks worth taking are the ones whose worst case fits on one page. Writing that page before you enter is what turns fear into a decision. Published 2025-10-04T00:00:00.000Z. 4 min read. Tags: letters, risk, action, opportunity, decision-making, growth.
- [Never Bet Your Life on One Door](https://rajdeepmondal.com/writing/never-bet-your-life-on-one-door): Summary: A second answer is what makes the first one negotiable. Build options on calm days, because a hard day leaves time to use them and none to make them. Published 2025-09-21T00:00:00.000Z. 5 min read. Tags: letters, strategy, resilience, decision-making, self-improvement.
- [Think for Yourself](https://rajdeepmondal.com/writing/think-for-yourself): Summary: Normal inputs produce normal results. The way out is a worldview you assemble from many fields and test before you trust it. Published 2025-09-15T00:00:00.000Z. 10 min read. Tags: mental-models, critical-thinking, self-improvement, first-principles.
- [Second Order Consequences](https://rajdeepmondal.com/writing/second-order-consequences): Summary: Every choice pays its first effect at once and its later effects for years. The skill is deciding before the two curves cross. Published 2025-09-09T00:00:00.000Z. 10 min read. Tags: decision-making, mental-models, long-term-thinking, time.
- [Deciding Is a Skill](https://rajdeepmondal.com/writing/making-decisions): Summary: Deciding is a trainable skill, and the training starts by cutting how many decisions reach you at all. Sort what is left by half-life and by the cost of undoing it. Published 2025-09-03T00:00:00.000Z. 6 min read. Tags: decision-making, mental-models, productivity.
- [Build Yourself First](https://rajdeepmondal.com/writing/build-yourself-first): Summary: The work pays out judgment, taste, and a standard you hold with nobody in the room. Money is the receipt, and it prints later. Published 2025-08-20T00:00:00.000Z. 5 min read. Tags: letters, discipline, self-improvement, motivation, mindset, diligence.
- [Stay Long Enough to Become Unavoidable](https://rajdeepmondal.com/writing/stay-long-enough-to-become-unavoidable): Summary: Staying past the point where the work stops answering is the whole edge. The minimum day is what makes staying survivable. Published 2025-08-14T00:00:00.000Z. 4 min read. Tags: letters, persistence, discipline, self-improvement, mindset, consistency.
- [The Will to Finish](https://rajdeepmondal.com/writing/will-to-finish): Summary: Finishing is a schedule you keep after the mood is gone. The test arrives on day two hundred, when the work is dull and nobody watches. Published 2025-08-08T00:00:00.000Z. 7 min read. Tags: discipline, focus, consistency, sacrifice, self-improvement.
- [Building Luck](https://rajdeepmondal.com/writing/building-luck): Summary: Luck is readiness meeting a moment you did not choose. Readiness is a system you maintain, and its leaks have names you can close this week. Published 2025-08-02T00:00:00.000Z. 4 min read. Tags: letters, self-improvement, action, motivation.
- [Luck Follows the Ones Who Move](https://rajdeepmondal.com/writing/luck-follows-the-ones-who-move): Summary: Courage is a rate, and you pay it in attempts made before you feel ready. Ten asks a month is the floor, and below that luck has nothing to land on. Published 2025-07-26T00:00:00.000Z. 5 min read. Tags: letters, courage, action, self-improvement, motivation.
- [No More Excuses](https://rajdeepmondal.com/writing/no-more-excuses): Summary: Your best excuses arrive as analysis. What beats them is a standard fixed before the day starts, and a count of the consecutive days it held. Published 2025-07-21T00:00:00.000Z. 5 min read. Tags: letters, action, motivation, self-improvement, discipline.
- [No One Is Coming to Save You](https://rajdeepmondal.com/writing/no-one-is-coming-to-save-you): Summary: Every standard you keep is enforced by you. The advantage that lasts is how you work across the stretch where nobody watches, and year five is where it shows. Published 2025-07-16T00:00:00.000Z. 4 min read. Tags: letters, discipline, self-improvement, motivation, mindset.
- [Start Today](https://rajdeepmondal.com/writing/stop-thinking-start-moving): Summary: Planning stops paying at the point where the next fact can only come from a move, and that move is almost always small enough to make today. Published 2025-07-11T00:00:00.000Z. 4 min read. Tags: letters, action, motivation, self-improvement.
- [Your Hour Has a Buyer](https://rajdeepmondal.com/writing/age-of-distraction): Summary: Companies measure your hour and sell it. The defense that works is one direction, protected long enough for the later effect to arrive. Published 2025-07-07T00:00:00.000Z. 5 min read. Tags: focus, productivity, mindfulness.
- [The Battlefield Moved](https://rajdeepmondal.com/writing/battlefield-moved-yc-ai-camp): Summary: The value moved from chat to finished work. The opening is the gap between what models already do and what shipped products do. Published 2025-06-19T12:00:00.000Z. 18 min read. Tags: ai, startups, agents, technology, founders.
- [Goechala: Sixty-Five Kilometers to One Sunrise](https://rajdeepmondal.com/writing/goechala-trek): Summary: Sixty-five kilometers, ten days and a 1 AM start bought one sunrise at 15,100 feet. Book Goechala when you want the route to be the hard part. Published 2025-05-04T12:00:00.000Z. 7 min read. Tags: trekking, travel, adventure, mountains.
- [Dayara Bugyal: The Trek I Was Not Fit For](https://rajdeepmondal.com/writing/dayara-bugyal-trek): Summary: I was not able to run one kilometer, then walked twenty-one over six monsoon days to Dayara Top at 11,830 ft. Book the trek first and train after. Published 2024-09-29T12:00:00.000Z. 9 min read. Tags: trekking, travel, adventure, mountains, beginnings.

## Contact and profiles

- [Contact form](https://rajdeepmondal.com/connect): Best route for specific collaborations and technical notes.
- [Calendar](https://cal.com/rajdeepmondal/30min): Scheduling link for focused calls.
- [GitHub profile](https://github.com/rajdeepmondaldotcom): Open-source work and repositories.
- [LinkedIn profile](https://www.linkedin.com/in/rajdeep-mondal/): Professional profile.
- [X profile](https://x.com/_rajdeepmondal): Short-form notes and public updates.