LinkedIn / case study

Building intelligence under live constraints.

An autonomous trading agent, observability layer and multi-agent development workflow built during the Claude Hackathon Madrid.

OutcomeTop 10 of 18 · 1,377 tests

Evidence leveldocumentedScale and architecture documented; adoption or savings are not overstated.

StackClaude Code · Python · agents · dashboards

01 / Context

The question behind the work.

Eighteen teams shared a live market where value, not activity, determined the score. The system had to make decisions under scarce actions, changing information and competing strategies.

02 / Built

Structure before decoration.

Contributed the intelligence layer, feed oracle, duel decision tree, dealer sell channel, Radio Rastro and four dashboard versions. The team used isolated branches and a fleet of Claude Code agents, with safety boundaries around the live game server.

03 / Result

What changed.

The team finished in the top 10 of 18, reached position 8, merged 38 pull requests in 30 hours and passed 1,377 tests. This is evidence of AI systems, observability and collaboration under pressure, separate from client employment work.

This case study is intentionally anonymised. It describes the shape, scale and outcome of the work without exposing internal data, client systems or confidential screenshots.

Continue

See the full LinkedIn practice.

Back to LinkedIn