Pranav Belhekar
I architect and ship AI systems that hold up beyond the demo.
I'm the co-founder and CTO of GrowthForge AI. Across 15+ delivered client systems, I've worked on agent orchestration, retrieval, memory, and the checks that make AI useful in production. I share what I learn in visual guides and my books.
Featured visual guide
A Visual Guide to AI Agent Memory
Your agent rebuilds its entire mind from scratch every single turn. How agents remember, forget, and what actually works in production — explained visually, from first principles.
Read the guide
My Books · ATLAS Volume I
Memory Management for AI Agents
Four connected chapters on how agents remember, lose context, forget, and revise what they believe.
Explore the book →Field notes in your inbox
Roughly twice a month. No spam, no LinkedIn-isms.
Selected client work
Work with me →-
Creative agency
Creative operations orchestration
A stateful multi-agent workflow from brief intake to delivery and invoicing, with retrieval from prior work, brand-compliance checks, and human approval before assets ship.
-
Customer communications
Multilingual PII detection
A production service that audits conversations for exposed personal data, including obfuscated text, with asynchronous processing, monitoring, and CRM write-back.
-
SaaS team
Enterprise knowledge assistant
A retrieval assistant grounded in the team’s Confluence knowledge base, with scheduled document sync and the ability to open Jira tickets from a conversation.
Open work you can inspect
ContextDoctor is my open source analyzer for RAG documents and chunks. It flags oversized chunks, duplicates, broken tables, and other problems before a model sees them.
I'm based in India, co-founded GrowthForge AI with Darshan, and spend most days either deep in a codebase or writing about why something I built last quarter doesn't work the way the demo suggested. Find me on X if you want to argue about agent architectures.