How wrishavsett/hsl works
HSL is a config-driven FastAPI invoice extraction service. A client uploads a PDF to /extract; the API validates and temporarily saves it, then the Gemini client loads extraction settings, rasterizes the PDF, and requests schema-constrained structured output from Gemini. The API resolves configured fields and returns JSON, while temporary files are cleaned up. The graph also retains the health check, shared error handling, and logging layer. Relationships follow the README and shown excerpts; unsampled implementation details are omitted.
Main entry points: api.py FastAPI routes · gemini_client.py Extraction orchestrator · helper.py Config validation · logger.py Logging
Components 16
HTTP ingress
- FastAPI routes
src/api.py - Upload validation
src/api.py - Temporary PDF
src/api.py - Field resolver
src/helper.py - JSON response
src/api.py
Extraction workflow
- Extraction orchestrator
src/gemini_client.py - Config validation
src/helper.py - Extraction settings
configs/extraction.json - PDF rasterization
src/helper.py - Rasterized pages
src/helper.py - Gemini client
src/gemini_client.py
Shared services
- Domain errors
src/error.py - Logging
src/logger.py
External dependencies
- PDF tools
- Google Gemini
Other components
- API caller
Connections 21
- API caller → FastAPI routes: sends request
README.md - FastAPI routes → Upload validation: routes upload
src/api.py - Upload validation → Temporary PDF: saves PDF
src/api.py - FastAPI routes → Extraction orchestrator: calls extraction
src/api.py - Extraction orchestrator → Config validation: loads settings
src/gemini_client.py - Config validation → Extraction settings: reads config
README.md - Extraction orchestrator → PDF rasterization: rasterizes PDF
src/gemini_client.py - PDF rasterization → PDF tools: uses Poppler
README.md - PDF rasterization → Rasterized pages: writes images
README.md - Extraction orchestrator → Gemini client: calls model
src/gemini_client.py - Gemini client → Rasterized pages: reads images
src/gemini_client.py - Gemini client → Google Gemini: sends prompt
src/gemini_client.py - Extraction orchestrator → Rasterized pages: cleans up
src/gemini_client.py - FastAPI routes → Field resolver: resolves fields
src/api.py - FastAPI routes → Extraction settings: loads field map
src/api.py - Field resolver → JSON response: produces data
src/api.py - FastAPI routes → JSON response: returns JSON
src/api.py - FastAPI routes → Domain errors: catches errors
src/api.py - FastAPI routes → Logging: logs events
src/api.py - Extraction orchestrator → Logging: logs events
src/gemini_client.py - Config validation → Logging: logs issues
src/helper.py