Challenge
Lender policy rules were fragmented across PDFs and tribal knowledge, forcing finance managers to manually interpret rate sheets during live deal structuring.
getBo
AI loan-structuring intelligence engine that turns lender policy documents into a conversational decision system for finance teams.
Project: AI lender intelligence engine
Role: Founder / Systems Builder
Status: Active development
Stack: Next.js · Supabase · OpenAI · Gemini
Approach
Built a pipeline that ingests each lender rate sheet, indexes only three routing signals into the database, then lets the model reason directly over the source document during inference.
Architecture / Stack
- Repo shape: JS/TS monorepo-style setup with two apps under apps/ (api + mobile), orchestrated via root package scripts
- Backend/API: Next.js (App Router) + React runtime in apps/api using route handlers as API endpoints (TypeScript)
- Mobile app: React Native + Expo in apps/mobile, with expo-router for navigation/routing
- Data layer: Supabase (supabase-js) with SQL migrations in supabase/migrations for Postgres-backed persistence
- LLM/AI: OpenAI (openai SDK) + Google Gemini (direct API integration for Loan Desk extraction and vision flows)
- State/UI tooling: TanStack React Query, React Navigation, AsyncStorage, SecureStore, NativeWind + Tailwind
- Build/tooling: TypeScript, ESLint (Next config), ts-node for server-side scripts such as voice-gateway
Outcome
Enabled finance teams to structure deals faster while grounding decisions directly in lender policy.
Impact
Improved consistency in deal decisions while reducing manual lookup time for complex structuring questions.