Doc AI
An AI service that accelerates form completion by extracting structured data from documents.
Where it started
Users manually transferred information from many document types into forms, creating delays and data-entry errors.
My part
Backend and AI Engineer
What I built
Using the Google Gemini API, I developed document ingestion, classification, multi-file analysis, structured JSON output and error handling. I integrated the service into form workflows used by production products.
Where it stands
Average processing time fell to about six seconds, while a representative form workflow dropped from 7–8 minutes to roughly 50 seconds.
What stands out
- 22 document categories
- Multi-file analysis
- Structured, verifiable output
- Reusable service across products
On the technical side
Using schema-constrained JSON, validation and explicit error paths instead of consuming raw LLM output made integrations more predictable.
Built with
DjangoDRFCeleryRedisGoogle Gemini APIPostgreSQL