Projects

Autonomous accounts-receivable agent with a three-tier autonomy gate.

Built at Cursor Hack London · 30 Apr 2026

Problem

Small UK businesses lose time and cash chasing late invoices, and letting an LLM act freely on financial decisions isn’t safe.

Built

• A FastAPI agent that chases overdue invoices, screens counterparties for AML risk and applies UK Late Payment Act interest.
• Every action passes a three-tier autonomy gate (AUTO / ESCALATE / HARD_BLOCK).
• Claude drafts the emails; the activity feed streams live over WebSockets behind JWT + bcrypt auth.

Stack

• FastAPI
• Claude API
• JWT
• bcrypt
• WebSockets
• SQLite
• React

Async code-review API: submit a GitHub PR or raw diff, get back structured findings (line, severity, issue, fix).

Problem

An AI first-pass review only helps if its output is structured and the API never blocks while the model runs.

Built

• FastAPI service with a typed Pydantic contract, the 202-Accepted async job pattern, input validation and a 7-test pytest suite.
• Claude review worker and Redis queue in progress.

Stack

• FastAPI
• Pydantic
• pytest
• Claude APIin progress
• Redisin progress

AI fitness-coaching bot on Telegram, grounded in your Strava data.

Problem

Generic training advice ignores the workouts you’ve actually done, and that data sits scattered across fitness apps.

Built

• A Telegram bot that syncs Strava activities through OAuth2 into Supabase.
• Free-text questions go to Claude with recent training as context, so coaching is grounded in real data.
• Dockerised and deployed on Render.

Stack

• Claude API
• Supabase
• Strava OAuth2
• TypeScript
• Docker
• Render

A REST API for exploring London’s blue plaques, built on Open Plaques data.

Problem

Open Plaques data is one flat list, so questions like ‘which London areas have the most plaques for scientists?’ are hard to answer.

Built

• A Flask + MongoDB REST API over the Open Plaques dataset, modelling each plaque as a nested document (subjects, location, categories).
• In progress: search and filter endpoints, and aggregation endpoints that compute counts and averages per area and category with MongoDB’s aggregation pipeline.

Stack

• Python
• Flask
• MongoDB
• Aggregation pipeline
• Open Plaques data