| What I built | Where | What it does |
| Key Account Intelligence agent (designed and built its core) | Alkemy | Researches a client's most important customers, finds who can introduce them, tracks progress and keeps the news up to date |
| Introductions agent | Alkemy | Works out who knows whom from shared employers and schools, and ranks who can make the best introduction |
| Lead-finding agent | Alkemy | Searches the web for companies that match a client's ideal customer; found 600+ in one search |
| Cost reduction | Alkemy | Cut the cost of each report by more than half by giving each agent only what it needs and testing a cheaper model |
| TradePilotAI prediction models | DevDash | Predict a stock's high, low and price swings for the day from 400+ signals, tested only on later dates so results are honest |
| TradePilotAI trade-check agents | DevDash | Check news, earnings, insider trades and the economy, and can stop a trade; a full run takes under 80 seconds |
| Plain English to database queries | Research | A T5 model that also reads the database layout; it beat a model without that on hard MongoDB queries |
| Crop recommendation from field sensors | IOE | Wireless sensors read the soil; a machine learning model suggests what to plant |
| Gift card platform for many businesses | Atrusta | Built end to end: Django REST Framework, React, Tailwind, PostgreSQL, Redis, Celery, live updates over WebSockets, Stripe payments, QR-code redemption, AWS S3, Mailgun email |
| Booking assistant for a skincare clinic | Atrusta | Books appointments for clients and adds them to the clinic's calendar |