Hackathon entries, products and systems. Each has an interactive diagram of how it works (illustrative, not live data). The five featured case studies are on the home page, and everything else is on GitHub ↗.
AI skin, hair and nail diagnosis. 7 specialised models across 4 platforms run in parallel with a majority-voting ensemble, returning results in about 3 seconds. Redis caching, a Hinglish voice assistant, PWA and Android. Built and deployed in a 36-hour hackathon.
Majority voting · click a model to flip its vote
M1
M2
M3
M4
M5
M6
M7
Illustrative Illustrative votes. The real system runs 7 specialised models across 4 platforms in parallel and uses a majority-voting ensemble.
Multi-tenant billing, isolation and rate limiting in one product.
A solo-built, multi-tenant SaaS on Telegram for job discovery, AI resume scoring and recruiter outreach. n8n pipelines with Groq Llama 4, Hunter.io, Apollo and the Gmail API on AWS EC2. Razorpay billing with webhook-driven payments, PostgreSQL tenant isolation and rate limiting.
One pipeline, separate tenant data
Telegramn8n pipelineGroq Llama 4 scoringHunter.io · Apollo · Gmail
Policy enforcement and infrastructure as code for LLM calls.
A policy-driven governance gateway for LLM APIs: Terraform-provisioned infrastructure, automated guardrail and policy checks, and a retry-aware orchestrator that enforces safety, routing and cost controls before requests reach the model.
Send a request through the gate
Request
Guardrail & policy checks
Retry-aware orchestrator
Model
Illustrative Illustrative request paths through the documented design.
An AI productivity coach that returns one concrete next action toward your goal, with reasoning, ordered steps and a time breakdown. FastAPI and Groq backend, Flutter and Firebase app, multi-language, plus an offline dino-style mini-game.
From a pile of options to one next action
Next action · one concrete step toward your goal Reasoning · why this step, now Steps · 1, 2, 3 in order Time · how long each part takes
Illustrative Illustrative. Real output: one next action with reasoning, ordered steps and a time breakdown.