academic-commercialization-agent

Academic Commercialization Assessment Agent is a Python/FastAPI application built with CrewAI that converts a research paper or topic into a structured commercialization assessment draft. It combines deterministic retrieval with six LLM stages—academic, patent, and market evidence specialists followed by writer, reviewer, and scorer—to produce source-linked reports with an auditable scorecard and risk notes. The project emphasizes controlled autonomy and evidence provenance. Retrieval is deterministic, sources are validated and registered, and scoring uses a deterministic weighted formula. Reports support research triage rather than formal due diligence. The system includes checkpoint recovery for interrupted runs, immutable recovery children, durable request receipts, and optional OpenTelemetry/OpenInference tracing. Deployment is via FastAPI with a build-free JavaScript client, Docker, and Railway. Access can be gated by operator-issued codes or bring-your-own-key credentials. The repository documents measured results, including a 30/30 end-to-end completion baseline across 10 topics and a 90-cell topology ablation, while clearly separating validated findings from historical experiments and unvalidated scoring policies.