How ashneetjha/finvision-ai works
FinVision AI is a document-digitization and invoice-audit web application. A client submits an invoice or selects a sample; the FastAPI orchestrator coordinates ingestion, OCR, table and financial extraction, signature detection, optional benchmarking, audit decisions, and report generation. Session workspaces retain inputs and reports, while the UI and download routes expose results. The README documents the overall workflow and capabilities; some internal wiring is supported by the sampled source index rather than complete app excerpts.
Main entry points: app.py FastAPI orchestrator · audit_agent.py Audit decisions · ocr_agent.py OCR perception · reporting_agent.py Workbook reporting
Components 14
Web and orchestration
- FastAPI orchestrator
src/app.py - Audit console
templates/index.html
Document extraction
- Ingestion
src/agents/ingestion_agent.py - OCR perception
src/agents/ocr_agent.py - Table reconstruction
src/agents/table_agent.py - Financial extraction
src/agents/financial_agent.py - Signature detection
src/agents/signature_agent.py
Validation and provenance
- Benchmark evaluation
src/agents/evaluation_agent.py - Audit decisions
src/agents/audit_agent.py - Traceability
src/agents/traceability.py
Reports and storage
- Workbook reporting
src/agents/reporting_agent.py - Session workspaces
data/sessions - Report output
data/output
Other components
- Client
Connections 25
- Client → FastAPI orchestrator: submits document
README.md - FastAPI orchestrator → Audit console: serves console
src/app.py - FastAPI orchestrator → Ingestion: ingests input
src/app.py - FastAPI orchestrator → OCR perception: runs OCR
README.md - FastAPI orchestrator → Table reconstruction: reconstructs tables
README.md - FastAPI orchestrator → Financial extraction: extracts fields
README.md - FastAPI orchestrator → Signature detection: checks authorization
README.md - FastAPI orchestrator → Benchmark evaluation: benchmarks extraction
README.md - FastAPI orchestrator → Audit decisions: audits document
README.md - FastAPI orchestrator → Traceability: builds provenance
README.md - FastAPI orchestrator → Workbook reporting: generates workbook
README.md - Ingestion → Session workspaces: stores session input
README.md - FastAPI orchestrator → Session workspaces: reads session files
src/app.py - Financial extraction → OCR perception: uses OCR tokens
src/agents/financial_agent.py - Table reconstruction → OCR perception: uses OCR tokens
src/agents/table_agent.py - Audit decisions → Financial extraction: checks extracted fields
src/agents/audit_agent.py - Audit decisions → Signature detection: checks signature result
src/agents/audit_agent.py - Audit decisions → OCR perception: checks confidence
src/agents/audit_agent.py - Audit decisions → Table reconstruction: uses table structure
src/agents/audit_agent.py - Traceability → Financial extraction: records field lineage
src/agents/traceability.py - Traceability → OCR perception: resolves source tokens
src/agents/traceability.py - Workbook reporting → Traceability: renders provenance
src/agents/reporting_agent.py - Workbook reporting → Session workspaces: writes session report
README.md - Workbook reporting → Report output: writes fallback report
src/agents/reporting_agent.py - FastAPI orchestrator → Client: returns results
README.md