/writing
Writing
Technical deep-dives on GenAI, multi-agent systems, and production infrastructure. Original essays, platform articles, and talks.
Essays
Original long-form pieces. Definitions written to be cited, not skimmed.
The Capability Lease
Long-lived credentials are the SQL injection of the agent era.
Part 1 of 3 in the Capability Lease series.
Read the essayLatest
Pulled from Medium and Substack in the last 30 days. Auto-synced on each build.
My Smog Map Counted 539 Fires. It Also Says Why That Number Is Low (opens on Substack in a new tab)
I turned an open-source 3D globe into Dhuan, a live view of India's smog season, in two evenings. Most of the design went into printing what the satellites miss.
296,000 stars, 45 credited clones, under 4,000 stars between them (opens on Substack in a new tab)
A dated ledger of the reverse-engineering wave, built from the repo, the issue tracker and the clone catalogue, with my method and what I could not verify.
My Agent Memory Database Has Five Ways to Forget. None of Them Erase the Trail (opens on Substack in a new tab)
In mnemo, forgetting is a signed event in a hash chain. What that buys (256 of 256 tamper checks caught), what it costs, and the retrieval result I can't claim yet.
Agent Leaderboards Reward Partial Credit. Production Does Not (opens on Medium in a new tab)
The top agent on Agents’ Last Exam scores 63.2% but fully finishes 38.2% of tasks. Here is the number I would gate a release on, and when the other one is fine.Agent Leaderboards Reward Partial Credit. Production Does No
Fifty-Seven Releases in Fourteen Weeks, Then I Froze My Own Project (opens on Substack in a new tab)
Provael kept shipping and its one real number kept getting older. Why the roadmap now holds three dated proofs and nothing else, and what the freeze costs me.
My Scanner Reads Every Skill Alone. The Attack Lives Between Two (opens on Medium in a new tab)
Splitting one malicious workflow across two agent skills made it work five times more often. Here is where I now put the check and the rule I use to decide.My Scanner Reads Every Skill Alone. The Attack Lives Between Two
Start here
New to my writing? Pick an entry point by interest area.
AI Development & Productivity
My viral article on building the ultimate Claude Code setup. 464 claps, real insights from daily usage.
Read: Ultimate Claude Code Setup (opens in a new tab)LLM Optimization
Technical exploration of model compression, quantization, and performance optimization techniques.
Read: AWQ Quantization (opens in a new tab)AI Career Strategy
A roadmap for building a career in AI, especially for young professionals in India.
Read: AI Career Roadmap (opens in a new tab)All articles
In-depth explorations of GenAI systems, infrastructure design, and engineering best practices.
View counts: author-side view count from Dev.to admin (not public reactions/claps).
I Spent Months Building the Ultimate Claude Code Setup. Here's What Actually Works. (opens on Medium in a new tab)
6 MCP servers, 130+ agents, 64 skills, and the truth about what's worth your time. No marketing fluff, just an honest assessment from real daily usage.
(dtwdistance) Dynamic Time Warping Distance (opens on Medium in a new tab)
Deep dive into Dynamic Time Warping distance concepts and practical use cases. Covers algorithm fundamentals, implementation patterns, and real-world applications in time series analysis.
AWQ: Revolutionary Approach to Quantization for LLM Compression (opens on Dev.to in a new tab)
Exploring Activation-aware Weight Quantization for efficient large language model compression and acceleration. A deep dive into cutting-edge optimization techniques.
Unleash the Power of Video-LLaMA: Revolutionizing Language Models (opens on Dev.to in a new tab)
Video-LLaMA brings video and audio understanding to large language models. Exploring how this breakthrough enables multimodal AI capabilities.
Perfscope: Measure-First Python Performance in the Age of Vibe Coding (opens on Medium in a new tab)
A lightweight, decorator-based profiler/APM that shows you exactly where time and memory go, on your laptop or in production.
Highlights of AI Index Report 2023 (opens on AI monks.io in a new tab)
Stanford University Human-centered AI (HAI) published the AI Index Report 2023. A comprehensive breakdown of key findings and industry trends.
Exploring TermGPT: Terminal-based Interactions with GPT (opens on Dev.to in a new tab)
A powerful tool for bringing GPT capabilities directly to your terminal. Command-line AI interactions made simple and efficient.
The Python Interview Almanac (opens on Dev.to in a new tab)
Comprehensive guide covering essential Python concepts, patterns, and best practices for technical interviews.
Introducing Shell-AI: Elevate Your Command Line with Natural Language (opens on Dev.to in a new tab)
Natural language interface for your command line. Execute complex commands using plain English instructions.
Jailbreaking GPT-4's Code Interpreter: Unleashing Untamed AI (opens on Dev.to in a new tab)
Exploring the boundaries and capabilities of GPT-4's Code Interpreter. A deep dive into its potential and limitations.
AutoDev vs Devin: The Pioneers of AI-Driven Software Engineering (opens on Medium in a new tab)
In-depth comparison of AutoDev and Devin, marking a new era where AI agents autonomously write, test, and deploy code.
AI Is an Elephant: A Simple Map of How We Got Here, and What Comes Next (opens on Medium in a new tab)
If you're a 20-25-year-old in India trying to build a career or business in AI, the hardest part isn't learning one more framework. A roadmap for understanding the AI landscape.
World Mental Health Day 2025: Small, Trustworthy Steps for Everyday Calm. Introducing MannSetu (opens on Medium in a new tab)
Today, Oct 10, the world spotlights mental health. Here's a simple way India can make it doable, one minute at a time. Introducing MannSetu AI wellness platform.
Footnote: llms.txt adoption sits at 10.13% with no measured ranking lift (SE Ranking, April 2026 (opens in a new tab)). I still ship one because well-formed agent-readable surfaces compound. See /AGENTS.md.
Writing topics
Areas I write about regularly, organized by category.
AI & Machine Learning
Python & Performance
India & Social Impact
Mental Health & Wellness
Events and community
Conferences and summits I've attended. Currently building toward speaking at events in 2026.
The Makers Summit by Inc42
India's largest product conference. Explored product management, growth strategies, and user-centric design with founders, PMs, and builders from across India.
RAISE2020, IIT Madras
Robotics and AI summit focused on cutting-edge research in autonomous systems, computer vision, and reinforcement learning.
IIT Madras AI Summit
Discussions on production ML systems, model deployment strategies, and AI ethics. Networked with ML engineers from leading tech companies.
The Production Agent
Weekly lessons from running 15+ AI agents in production
Interested in AI/ML content?
Follow my writing on Medium and Dev.to for technical deep-dives, case studies, and practical insights from building production AI systems.