Notes from building
Write-ups on RAG, agentic AI, MCP, and deep learning — what worked, what broke, and why.
-
2026-03-18 · 8 min read
Building Saarthi AI: giving a commute-planning agent memory with MongoDB MCP
Notes from building Saarthi AI at the Google Cloud Rapid Agent Hackathon — why we kept the risk score deterministic, what MongoDB MCP actually bought us over a normal database call, and what broke during the demo.
-
2026-01-25 · 7 min read
Grounded RAG: how Scholar_Mate AI avoids hallucinating on your own notes
Building a RAG assistant for real student notes meant handling scans and handwriting, not just clean PDFs — and making it comfortable saying 'I don't know' instead of guessing.
-
2025-09-10 · 6 min read
From 66.7% to 90.1%: fine-tuning ResNet50 for pneumonia detection
A baseline ResNet50 barely beat guessing the majority class. Here's what actually moved accuracy 23 points, and why Grad-CAM mattered more than the accuracy number itself.
-
2026-04-02 · 6 min read
What three AI hackathons taught me about shipping instead of demoing
Between the Google Cloud Rapid Agent Hackathon, the Redrob Data & AI Challenge, and a couple of smaller builds, a few lessons kept repeating — mostly about what breaks the moment judges start clicking around instead of watching a script run.