Inspiration Manual amendment review in banking is slow, inconsistent, and easy to miss critical risk changes. We wanted a tool that could translate dense legal language into clear, business‑focused insights for non‑technical judges and credit teams. The goal was demo‑first: show immediate, credible value in under three minutes.
What it does RiskDiff compares an original loan agreement with its amended/restated version, detects added/removed/modified clauses, and explains the commercial impact in plain English. It highlights risk level, who benefits (borrower/lender/neutral), and lets users filter to the most material changes.
How we built it We built a desktop‑first web app with a simple but effective pipeline: Parse documents into structured clauses using heading/number heuristics. Compare clauses semantically to detect added/removed/modified changes. Explain impact in plain English with risk level + beneficiary tags. Visualize results in a clean fintech UI with filters and drill‑downs. Frontend: React + TypeScript Backend: Node + Express AI: Ollama (free local) with a rule‑based fallback Challenges we ran into Document variability: legal agreements don’t follow a perfect structure, so parsing needed to be robust without overfitting. Semantic diffing: we tuned similarity scoring to reduce false positives while still catching meaningful changes. Demo performance: we kept everything fast and smooth so the UI stays responsive. AI cost/availability: we added a free local AI path (Ollama) so the app remains smart without paid APIs.
Accomplishments that we're proud of Built a clear, judge‑friendly UI that turns dense legal text into business insight. Delivered end‑to‑end analysis: parsing → diffing → explanations → dashboards. Added free AI support via Ollama so the project is usable without paid services. Produced realistic mock agreements for a strong demo.
What we learned Even imperfect parsing is valuable if the output is clear, risk‑focused, and explainable. Information hierarchy (summary → filters → details) matters more than raw model accuracy in a demo. AI is most useful when it reduces jargon and highlights impact quickly.
What's next for RiskDiff Exportable diff reports for audit and compliance workflows. Multi‑amendment comparisons across a loan’s lifecycle. Institution‑specific risk scoring rules. Production deployment with secure document handling and role‑based access.
Built With
- express.js
- framer
- javascript
- lucide-icons
- multer
- node.js
- ollama-(optional)
- openai-api-(optional)
- pdf-parse
- radix-ui
- react
- shadcn/ui
- tailwind-css
- typescript
- vite
Log in or sign up for Devpost to join the conversation.