Lalit Shewani

Product manager by day.Builder by night.

My notes on how AI products really get built, from the technology to the people who lead and use them.

Product lessons

Stuck on something that isn't here? Tell me about it. It may become the next lesson.

How the tech really works

Coming soon

The first notes are on their way

How the models, retrieval and agents under AI products actually work, written for product managers.

Late-night builds

LangGraph · RAG · Agents

Compass.ai

Problem
Skills fade when you stop using them, and most learning roadmaps don’t adapt to you.
Approach
A multi-agent setup in LangGraph, with retrieval (RAG), that generates the curriculum as you learn.
Outcome
Still in progress. A first pilot is planned for early 2027, with product managers moving into AI product management.

Ollama · Local AI · macOS

Privacy-first file organiser

Problem
Organising files with AI usually means sending them to someone else’s server.
Approach
An AI model that runs on your own Mac through Ollama, with batching and parallel processing.
Outcome
A 10x performance improvement, and your files never leave the laptop.

A bit about me

I'm Lalit. I work on AI products at Ford, and spent the eight years before that at JP Morgan Chase.

Today that means document intelligence, an audit agent that keeps a person in charge, and AI tools for teams outside engineering. At the bank I went from writing trade-processing code to leading data and machine learning products, and helped run the Bengaluru PM community events.

Lalit Shewani, arms folded, looking to the side

8+years across product and engineering

  1. 2018Software Development Engineer IIJP Morgan Chase
  2. 2021Associate Technical Product Manager, Data PlatformJP Morgan Chase
  3. 2023Technical Product Manager (AVP)JP Morgan Chase
  4. 2026 · nowSenior Product Manager, AI Advancement CenterFord