Anonymous builder1 day agoJudging locked: Event build

Team 20

North Star

North Star is an autonomous growth PM agent that finds where users drop off, designs and launches A/B experiments in Statsig, automatically kills variants that hurt revenue, and ships the winners with a full audit trail.

Check your own project

Community votes recognize the build and do not affect official judging.

Review the project

Start with the source code, then open the demo or video if available.

View GitHub repository
Project description
Growth teams run maybe one experiment a week. North Star runs the whole experimentation loop on its own: observe, diagnose, design, launch, protect revenue, decide, and learn. It reads funnel data, finds the leakiest step (split by mobile vs desktop), recalls lessons from past experiments, and designs A/B variants as parameter changes, so nothing needs a deploy. A human approves the launch, then North Star creates and starts the experiment in Statsig. While it runs, a revenue guardrail watches every variant: in our demo, a stakeholder's "hide pricing" variant lifted signups but cut revenue per user by 65%, and North Star killed it in Statsig automatically. When a winner is clearly significant, it ships it in Statsig and writes a decision memo with the numbers. The model makes the judgment calls (diagnosis, experiment design, memo). Code enforces the safety (significance tests, minimum samples, guardrails). Every step lands in an audit trail with the data, model output, and cost behind it. Built with: Superlinked SIE Serverless (all agent reasoning, Qwen3.8 27B with strict JSON-schema output), Statsig Console API (create, start, kill, ship), Amplitude (live event stream), SingleStore (events, experiment registry, audit trail), and Cognee Cloud (experiment-memory knowledge graph). A full experiment cycle costs about half a cent in model calls. Traffic in the demo is simulated; assignment, tracking, statistics, and every agent decision are real.
Tools used