Member story · Cohort 08 · August 2026

Risk, from the inside. Decks and recordings, at his own pace

AVP at JPMorgan Chase in Maharashtra, in risk management and business consulting, after Credit Suisse. An engineer in electronics and telecommunication first, then a master’s in finance.

Vinit’s portfolio

Vinit Patil AVP, Risk Management at JPMorgan Chase

Vinit’s feedback

In Vinit’s own words

What Vinit wrote about the course after Cohort 08. These are real people with real portfolios, not marketing copy. Want to talk to Vinit? I’ll intro you.

Vinit PatilCohort 08 · August 2026

Vinit on the course

Overall, I had a good experience with Curious PM. The content was good.

Ankit has kindly shared all the decks and session recordings, so I can revisit them at my own pace. I plan to go through it all again once I have more bandwidth. Ankit and the team are also always available to connect and help with queries, which I really appreciate.

I also like that the team keeps experimenting with new AI tools and platforms and sharing new content with members, both during and after the 8 weekends.

✓✓

What changed for Vinit

Every deck and recording, at his own pace.

A team always available for queries.

New AI tools shared, during and after.

Vinit learned alongside PMs from

The path

Engineering, a master’s in finance, then banking

2013
Summer Intern
State Bank of India
Then
Risk Management
Credit Suisse
Now
AVP, Risk Management
JPMorgan Chase
Aug 2026
Curious PM
Cohort 08, August 2026

The work

What Vinit built, and writes

See everything on Vinit’s portfolio ↗

01 / 04

Database-connected AI chatbot with evals

A chatbot built in five stages, from a plain OpenAI call to one with conversation memory, tools and a live database behind it, then every answer traced in Langfuse and scored by AI judges.

OpenAI Streamlit PostgreSQL Langfuse

Open ↗

What it teaches a PM

Memory was the smallest change with the largest effect; evals made the answers measurable.

02 / 04

Podcast search engine and MCP server

Semantic search over a thousand podcast episodes, then the same engine exposed over the Model Context Protocol so an AI assistant can search, keep notes and run research briefs.

MCP RAG Pinecone FastMCP

Open ↗

What it teaches a PM

An MCP endpoint is a live API: ship it with auth.

03 / 04

Podcast RAG and fine-tuning

Question answering over 149 podcast transcripts, whole episodes and 900-character chunks side by side, then a small open model fine-tuned with LoRA to speak in a chosen persona.

RAG Pinecone Claude Fine-Tuning

Open ↗

What it teaches a PM

Retrieval for facts the model has never seen, fine-tuning for how it speaks.

04 / 04

Personal AI agent with voice

One agent loop that searches and reads the web, remembers you between sessions, runs a job search from your resume and answers the phone, with chat and voice sharing one memory.

AI Agents mem0 ElevenLabs Twilio

Open ↗

What it teaches a PM

The assistant grows by registering tools, not by adding branches.

What Vinit writes

vinitpatil.curious.pm3 min read ↗

A Risk You Can Explain Is a Risk You Can Manage

Models, reports, dashboards and limits: none of it helps if the people making decisions cannot say, in plain words, what could go wrong and why it matters.

vinitpatil.curious.pm3 min read ↗