AI is magic, and I'm the wizard 
Somewhere between raag and RAG.
Proof of work
Threadline
Most company-knowledge AI runs two engines: one to search, one to answer. Threadline runs one that does both, so every question costs less, and the same map lets your team find simple answers without asking the AI at all.
Read the case study →Vatavaran
A normal AC reacts after you're already uncomfortable. Vatavaran sees it coming: it forecasts how your room will feel, understands plain requests like "it's a bit chilly", and adjusts before you notice.
Read the case study →NeuroSense
Most medical AI hands you a verdict and asks for trust. NeuroSense shows exactly what it looked at in a patient's handwriting.
Read the case study →NeuroSense
One shaky spiral could mean many things. A doctor wouldn't stop at one test, so NeuroSense doesn't either: it checks the voice and a brain scan too, and looks for all three to agree.
Read the case study →AutoML Studio
Let the data choose its own path.
Which model, how to clean the data, which settings: the answers were already textbook rules. So I turned the data into numbers and gave them a pathway, where a rule or a head-to-head test at every fork picks the best-suited method. I'm just the mediator.
Read the case study →Experience
- Built an incremental OHLC data pipeline (Playwright) syncing live market data for NIFTY 50 stocks, storing full history locally to minimise redundant fetches
- Implemented a full candlestick pattern library (Doji, Three Soldiers, Evening Star and more) with cross-stock pattern search
- Architected toward a notification and auto-trade system for confirmed pattern signals
- Built and evaluated two architectures for a RAG knowledge system over meeting transcripts, GitHub and Slack, with LLM-driven tree traversal and automatic ClickUp task generation
- Wrote BDD test suites and built a resume parser that normalises any resume into structured JSON for an HR dashboard
- Built AI internal tools (Gmail email classifier, Telegram inbox bot) with Hermes Agent
- Analysed the company's stock, purchase and sales records to find which products keep moving and which quietly stall
- Turned those patterns into buying advice, so working capital stops getting frozen in stock nobody is asking for
- Also designed and built the company's customer-facing website · visit live →
Side quests
Smaller builds I made because I wanted them to exist.
Never Have I Ever: Crazy Mode
A party game with a snarky AI host. Best played in person, loudly.
Play it live →Counting Station
Hardware shops still count nuts and bolts by hand, one at a time. Counting Station lets a shopkeeper tip a handful onto a mat and get the count from a camera instantly, and it speaks up when it isn't sure.
See how it works →Candlestick Screener
Reads stock price data and automatically spots common candlestick patterns.
GitHub →
