Nova DueDiligence

• 

  Interface

• 

  Dashboard With Results

• 

  Deep Analysis Showcased

• 

  Sample Comparison Between a High-Risk Contract and SEC Filing (Found Online)

• 

  Document Q&A Sample

• 

  Sample Report Response

Nova DueDiligence draws from real-world pain points in M&A due diligence for its hackathon pitch. Inspiration

Inspiration

Mid-market PE funds and corporate teams drown in 50-200 document reviews (contracts, NDAs), taking 3-5 days at $400-800/hour—Deloitte's 2023 survey flags this as a top deal failure cause due to manual processes, fatigue, and missed cross-doc conflicts.

What it does

Uploads PDFs for parallel 4-agent analysis (Risk, Financial, Obligations, Compliance) via Nova Pro: extracts structured risks/flags with confidence scores, 0-100 risk rating, semantic QA/RAG (Nova embeddings), cross-doc comparison, risk dashboard, and PDF memos—all in 10-18s vs. hours manually.

How we built it

Single-file React/HTML frontend (Chart.js, jsPDF); FastAPI backend (Python 3.11, pypdf); ThreadPoolExecutor for parallel Nova Pro agents (300K context, JSON schemas); in-memory cosine sim vector store with auto-fallbacks.

Challenges we ran into

JSON markdown fences (fixed: regex/dual-parser); AWS rate limits (auto-sequential embed + UI warnings); scanned PDFs (graceful empty-text); agent crashes (defaults cf=0.5); Hallucinations (tighter prompts + cf triage).

Accomplishments that we're proud of

Parallelism (10-18s vs. 40-60s sequential, ~4x speedup); auditable traces/confidence scores; human-in-loop safeguards; $0.04-0.08/doc cost (vs. $150+/hr labor); fits existing workflows for mid-market PE/corporates.

What we learned

Parallelism via ThreadPoolExecutor bounds time to slowest agent; Nova Pro excels at 300K-token legal docs/JSON schemas; enterprise AI needs cf-scores for triage, full traces for audits, transparent fallbacks—no silent failures.

What's next for Nova DueDiligence

Docker/ECS deploy; DB persistence (RDS); OCR (Textract); fine-tune on legal datasets for 95%+ cf on critical flags.

Built With

• aws-bedrock
• bedrock
• bedrock-converse-api
• chart.js
• fastapi
• html/js
• in-memory-vector-store
• invokemodel
• jspdf
• pypdf
• python-3.11
• python-dotenv
• react
• threadpoolexecutor