At NexHacks, a team built Roly Poly to simplify prediction market analysis using Polymarket data. The tool addresses a core problem: users struggle to translate prices into actionable risk metrics like expected value, Kelly-optimal sizing, and Value at Risk. Their goal was creating a clear interface showing +EV or -EV for simulated bets, making probabilistic reasoning more accessible.

The development hit a major coordination wall. Integrating sponsor tools like numeric reasoning and observability systems at 4 AM proved chaotic. Every change required touching multiple components, making debugging a guessing game. The team realized the core risk wasn't technical performance but managing parallel workflows under hackathon time pressure.

They shifted by treating DevSwarm as an orchestration backbone. Instead of monolithic endpoints, they built a runtime agent pipeline: a context builder, a compressor, a tutor, and an evaluator. Each specialized agent ran independently with structured handoffs. This parallelized development, with six builder agents working on dedicated git branches for backend, reasoning, evaluation, and UI.

The final product included a 3D risk visualization, a numeric engine for EV and VaR calculations, live Polymarket data, and an automated evaluation loop using Arize Phoenix. By organizing complexity into parallel, specialized agents, the team finished a cohesive demo in 24 hours, placing 3rd overall and demonstrating a new model for building AI-heavy products under pressure.