🔬 AI-Powered NSF Grant Matching Engine

End-to-end system that parses NSF calls, scores researcher fit, and assembles “dream teams” in minutes instead of weeks.

Key Outcomes

• 78% reduction in team assembly time; 23% uplift in skill coverage vs. manual picks.
• 100% accuracy on eligibility checks by combining rule parsing + semantic filters.
• Production Streamlit front-end so research offices can search solicitations and export briefs instantly.

System Architecture

What Runs Under the Hood

• TF‑IDF + dense embeddings blend to score expertise, with logarithmic boosts for past NSF wins.
• Constraint engine checks PI eligibility, institutional caps, and collaboration history before finalizing rosters.
• LLM-powered gap analysis crafts stakeholder-ready reports highlighting risks, mitigations, and go/no-go.

Delivery & Ops

• Python + Streamlit app deployable on campus infrastructure; uses uv/poetry for reproducible environments.
• Pydantic data contracts and pytest suite guard parsing edge cases across PDF formats.
• Tutorials walk grant offices through one-click ingest, team review, and export workflows.