A personal engineering journalest. Chennai · based SF

Field notes · on tools, agents & the people who use them

Product engineer building AI tools for real workflows.

I build product surfaces, agent infrastructure, and local-first AI tools across developer workflows, internal platforms, and web3-adjacent networks.

Write to me Stealth work

Current work

two desks, one obsession

01 / Metacomp Technologies

Product engineering for specialized teams.

Frontend product development, backend workflows for mesh generation software, customer support, documentation, and new product exploration.

02 / rootnode labs

Vertical agents for focused operators.

Focused agent systems with RAG pipelines, MCP integrations, fine-tuning loops, and domain-specific interfaces for complex work.

What I work on

Product systems

Frontend, backend, docs, support, and customer feedback loops for engineering software.

Agent tooling

MCP servers, RAG pipelines, local model workflows, and vertical assistants with real inputs.

Domain workflows

Interfaces and automation for specialized teams, messy data, and high-friction internal processes.

Index of works

shipped systems / prototypes / infrastructure

No.
01

Obsidian Local LLM Helper

Local AI tool

An Obsidian plugin that connects notes to locally running models for writing assistance, summarization, and organization without sending vault data to a cloud service.

Keeps private writing workflows inside the user vault.

Obsidian / Ollama / LM Studio

Open sourceopen →

No.
02

GitInformed

AI news system

An AI-powered RSS aggregator that curates and summarizes developer news across sources, with models filtering for relevance before anything reaches the reader.

Turns noisy developer feeds into a smaller set of useful signals.

Next.js / LLM filtering / RSS

Liveopen →

No.
03

WaiverBuddy

Sports decision aid

A fantasy football assistant that compares player usage, matchup difficulty, and projections to surface better waiver wire pickups.

Helps fantasy managers make faster waiver decisions each week.

NFL data / projections / matchup analysis

Liveopen →

No.
04

Lenny's Council

Grounded expert panel

A local-first AI expert panel over Lenny Rachitsky's podcast and newsletter corpus. It selects relevant experts, runs multi-round discussions, and cites claims back to the source chunks.

Turns 295 podcast episodes and 349 newsletter essays into a cited advisory board for product questions.

SvelteKit / Postgres / pgvector / Ollama / Bun

Open sourceopen →

No.
05

mStack

Product messaging workflow

A product communication workflow for coding agents. It helps founders and product teams collect context, choose the strongest angle, draft launch updates, critique weak copy, and package channel-specific assets.

Turns messy release context into a structured launch loop: context, angle, draft, critique, package, retro, remember.

Codex / Claude Code / agent skills

Open sourceopen →

No.
06

Sidekick

Browser AI

A Chrome extension for local AI assistance in the browser: summarize pages, rewrite text, and ask questions about visible content using local models.

Brings private AI actions directly into day-to-day browsing.

Extension APIs / local models / page context

Chrome extensionopen →

No.
07

TheVerge MCP Server

Agent infrastructure

A Model Context Protocol server that fetches and parses recent articles from The Verge for assistants that need live product and industry context.

Gives MCP-compatible assistants current tech coverage.

MCP / feeds / parsing

Open sourceopen →

No.
08

Farcaster MCP Server

Social protocol tooling

An MCP server for Farcaster that can pull casts from users, search channels, and expose protocol activity to AI assistants through a stable tool layer.

Turns decentralized social context into agent-readable tools.

MCP / Farcaster / channel search