Hackathon · 2026 · Tokens LDN Multi-Agent Hackathon · London 2026
Basket
A team of five AI agents that spots product reformulations by watching complaints on the web, so brands find out in weeks instead of waiting for quarterly sales data.
Track Winner - Tokens LDN Multi-Agent Hackathon 2026
First place on our track at the Tokens LDN Multi-Agent Hackathon in London, 2026.
Python · FastAPI · Next.js · ClickHouse
Challenge
When a brand quietly changes a recipe, the backlash lands on the web within days but only shows up in sales data months later. By the time a category manager sees it in a quarterly review, the damage is done.
Approach
Basket is a pipeline of five agents run by a FastAPI orchestrator. Give it a product name and it finds the reformulation date, pulls complaints from news and the web, classifies them with traceable rules, rolls them up week by week in ClickHouse, and publishes a sourced alert the moment complaints spike.
What it does
- Five-Agent Pipeline. A date finder, a retriever, a classifier, an aggregator, and a publisher, each with one job, coordinated by an orchestrator.
- Finds the Turning Point. ClickHouse rolls complaints up by week and flags the inflection where a spike starts.
- Every Claim Has a Source. Complaints are classified with traceable rules and every published alert links back to the original mentions.
- Survives Demo Day. Local fallbacks kick in if any external service goes down, so the pipeline keeps running.
Gallery
Photos
