Mandi Mind
A lightweight AI application built around Gemma 4 and real Government of India agricultural market data.
Hi, I'm
I build software and write about it.
Currently working on Gemma 4 Developer Agent Competition.

Contributed directly to the official Kaggle CLI. Fixed code-competition submission validation architecture by adding API-level input checks and regression tests, preventing invalid requests from reaching the backend.
Built scalable retail data solutions processing pricing and availability across 1,000+ North American locations. Designed automated ETL pipelines using n8n and Python, and migrated product data to PostgreSQL for exploratory analysis.
A lightweight AI application built around Gemma 4 and real Government of India agricultural market data.
A Retrieval Augmented Generation system using LangChain, ChromaDB, and Groq, deployed via Streamlit.
Completed all 4 milestones to achieve Completionist++, ranking in the top ~514 global finishers out of 115k+ participants.
Global Open-Source AI Celebration by MLH, DEV & DigitalOcean
Hacktoberfest 2026 pivoted from counting pull requests to hands-on open-source AI and community participation. Rather than submitting generic PRs, participants unlocked virtual stickers and milestone tiers by building weekly AI challenges with open-weight models (Google Gemma 3 & Prior Labs TabPFN), participating in technical livestreams, and actively contributing across Discord sessions.
M1
M2
M3
M4Local-first freelance billing engine and revenue defense copilot. Pairs local Google Gemma (via Ollama) to synthesize professional deliverables with Prior Labs' TabPFN-3.5 foundation model to predict budget overruns and scope creep from historical sessions. Enforces deterministic 6-min/15-min agency rounding.
Offline edge AI backcountry expedition companion engineered to operate in airplane mode with zero cellular connectivity. Employs Gemma 3 (1B/4B local Ollama) for sensory landmark navigation and Wilderness First Responder (WFR) emergency triage, combined with NOAA solar calculations and local Web Speech audio briefings.
Private, local-first recipe cookbook that transforms rambling Hindi/Hinglish audio voice memos into structured, searchable family recipes. Uses faster-whisper (int8 quantized) for local speech recognition, Gemma 3 (4B via Ollama) for JSON extraction, and local SQLite for storage.
Reflections on my very first Hacktoberfest: building 3 open-source AI projects with Google Gemma and TabPFN, reaching Milestone 4 (top ~0.45%), and key lessons learned.
Read post →Stuck thinking how to start open source contributions? Learn how I find and approach existing open source projects that accepts my pull requests.
Read post →How I got my first pull request merged as an open source contributor in kaggle-cli.
Read post →