Built, as a freelancer for a coaching institute, an exam-paper generator now used by 1,200+ students: LLM-written questions go through a Dockerised LaTeX service that outputs print-ready English or Gujarati papers.
Projects
Things I've built: full-stack apps, AI systems, and the odd weekend experiment. Most are open source.
Self-hosted RAG over PDFs, scans, audio and video in 5 independently scalable services. Retrieval benchmarked stage by stage on BEIR SciFact (nDCG@10 0.706 to 0.724), duplicate chunks cut from 17% of top-5 results to 0%, and crash-safe Redis Streams workers.
Shipped an asset-management system with role-based access for 4 roles across 20 tables, a Ctrl+K search over 4 entity types, and real-time notifications over SSE and Redis that replay anything missed on reconnect.
Ranked 94th of 19,900+ registrations at the Amazon ML Challenge 2026 by matching 10 million noisy business records: multi-search retrieval that keeps 99% of true pairs, a LightGBM pair scorer, and an expected-F0.5 decision rule.
Made retries safe in a self-hostable job tracker, with zero double-created records, by adding Redis idempotency keys and rate limits to a versioned Spring Boot REST API with revocable sessions.
Let field sales teams capture leads with no connection by building an offline-first CRM that syncs later, transcribes calls with Whisper, scores sentiment and schedules Google Meet follow-ups.
Work in progress: a real-time multiplayer race through Wikipedia links for up to 8 players, with live leaderboards pushed over Server-Sent Events.
Won the Adani Powermind Hackathon as a team with an energy-demand forecasting console (P10-P90 bands, anomaly alerts, retraining). I built the LightGBM forecaster: 2.59% day-ahead MAPE on a held-out year, 30% below the naive baseline, with no look-ahead features.