M&A Assessment Accelerator

"M&A Assessment Accelerator is one of the most technically complete and domain-credible submissions in this hackathon. Built solo in 35 hours, it compresses a 2-week M&A target assessment into 2-4 hours using 9 specialized Band agents across three phases that mirror how real deal teams operate. The architecture is genuinely multi-agent: Phase 1 runs 6 parallel research agents (buy-side and sell-side country, sector, and company profiles), Phase 2 has the @risk-agent synthesize all 6 outputs into a cross-workstream risk matrix using DeepSeek-R1, and Phase 3 produces a Go/No-Go recommendation with valuation range via Gemini. Examining the risk_agent code directly, the Band SDK integration is proper — band.on_message() with each agent running as a distinct identity, structured Pydantic schema validation (RiskAssessment.model_validate_json), hard business rules enforced in code (RED rating automatically sets human_review_required=True), and a graceful fallback to standalone test mode when Band SDK is unavailable. The 18-commit history shows a disciplined build progression: shared infrastructure first, then sell-side agents, buy-side agents, synthesis agents, coordinator, FastAPI backend, real-time SSE dashboard, demo data, and reliability fixes. The disk-based SHA256 response cache, swappable model config via models.yaml, and pre-cached demo scenarios all show production thinking. The model tiering is smart: Gemma 4 31B (free Google AI Studio) for orchestration and deal rationale, Qwen3-14B (Featherless) for research agents, DeepSeek-R1 (Featherless) for risk reasoning. This keeps costs near-zero while using the best model for each job. Minor gaps: the README live demo URL is still a placeholder (YOUR_DEPLOYED_URL), and the deployed demo at Render may have cold-start latency. The note about flagged GitHub org in the viewer is a viewer-side artifact from the judge's own organization — the repo itself is clean under ryker-code personal account with no issues."