Finova AI is an AI-powered document intelligence platform for loan verification and KYC workflows. It helps financial teams turn uploaded customer documents into structured, reviewable, and validated data using OCR, computer vision, and LLM-based extraction.
The product is designed around a practical intake-to-review flow rather than a generic OCR demo. The goal is to reduce manual verification time while keeping human reviewers in control of the final decision.
Loan and KYC workflows are still heavily document-driven. ID cards, payslips, and bank statements often arrive in inconsistent formats and require repetitive manual review before they can support a lending decision.
Finova addresses that problem by combining document preprocessing, OCR, structured extraction, validation, and reviewer-facing verification into one platform. The result is a pipeline that is faster than manual handling, but still auditable and operationally trustworthy.
- Accepts customer document uploads for verification workflows
- Normalizes PDF and image inputs into a consistent page-processing pipeline
- Uses OCR and computer vision to extract layout-aware text
- Converts unstructured content into structured JSON fields
- Runs validation rules to surface inconsistencies and missing information
- Supports human-in-the-loop review with persisted artifacts and review signals
The current MVP scope is intentionally narrow:
- Supported document types:
id_cardpayslipbank_statement
- Supported file formats:
.jpg.jpeg.png.pdf
- Architecture baseline:
- modular monolith
- local Docker Compose development
- cloud-oriented deployment design on Alibaba Cloud
- Frontend: Next.js
- Backend: FastAPI
- OCR: PaddleOCR
- Image processing: OpenCV
- Structured extraction: LLM with JSON-only output
- Database: PostgreSQL
- Object storage: MinIO for local development, OSS in cloud architecture views
- Cloud architecture target: Alibaba Cloud
The cloud documentation is organized as a multi-diagram set so each architectural concern can be understood independently without turning one image into a wall of lines.
High-level view of the production footprint on Alibaba Cloud, including edge entry, application services, worker tier, and core managed services.
The core processing path from synchronous upload requests into asynchronous OCR, extraction, and validation.
A data placement view covering transactional state, raw uploads, page artifacts, OCR output, extraction artifacts, backup, and encryption boundaries.
The north-south request path through DNS, CDN, WAF, Cloud Firewall, public entry, and user-facing runtime.
The security control view for protected runtime, KMS, Security Center, audit logging, bastion access, and protected data services.
The build and delivery view covering source control, pipeline, registry, infrastructure orchestration, and deployment targets.
frontend/- Next.js application and review interfacebackend/- FastAPI application, pipeline logic, persistence, and testsdocs/- architecture assets, agent docs, and supporting project documentation
Finova AI is built to make document-heavy financial workflows faster, more consistent, and easier to trust. It combines automation with reviewability, so teams can accelerate verification without giving up operational control.