arnav7t.

About
Work
Experience
Writing
Contact

resume ↗

Writing

Things I've
worked out in words.

Essays on AI systems, engineering decisions, and the occasional rabbit hole that turned out to matter.

You Don't Need Multi-Agent Systems (Until You Really Do)

Most agentic workflows can be solved with a single ReAct loop and structured tools. Multi-agent architecture makes sense for two specific reasons: context hygiene and permission isolation.

AI Engineering Agents

A Greeting Shouldn't Cost an Embedding

While building an air-gapped RAG chatbot, I realized the most important optimization wasn't retrieval or prompting — it was deciding when none of them were needed.

RAG ML Optimization

A Dig into TOTP

A deep dive into how Time-based One-Time Passwords actually work — from the HMAC-SHA1 internals to the dynamic truncation that makes it tick.

Security Cryptography

Building RecapGen

How I built a meeting summarisation tool — the architecture decisions, the AI pipeline, and what I'd do differently the second time around.

hookd RecapGen

© 2026 Arnav Thakur