Hi, I'm Mahesh.

I'm building things for the web. This is my personal space where I log my thoughts, notes, and updates. Welcome to my digital garden.

• Jul 31, 2026

  Your LLM should never see a secret

  Keeping a credential out of an LLM's context is necessary but not sufficient. The design I am building toward, the boundary that already holds, and the harder one that is still open: keeping the secret away from code the model can run.

• Jul 25, 2026

  Making Multi-Agent Orchestration Survive a Crash

  Why long-running agent tasks die on restart, and how to model delegation as a durable state machine on a message bus: one persisted event per transition, no polling, resume from the last recorded state.

• Jul 20, 2026

  Why agent memory should be an append-only log

  Mutable agent memory loses provenance and makes deletion a hope. Modeling memory as an append-only log where state is a fold that reads only the log makes erasure structural: remove the source events, re-fold, and nothing survives.

• Jan 5, 2026

  Upskilling System

  High Level Design practice/interview with AI.

• Dec 27, 2025

  Authentication Resources

  Links to authentication libraries and services.

• Dec 27, 2025

  Blog Generated Podcast

  Idea: Text to speech using gemma 3 for blog pages with emotions auto injected.