On-Device Document Extraction

The idea was to test and demo a Local LLM extracting and processing sensitive data opposed to a cloud LLM. This way PII data never the local environment and organizations can rest assured that their data is secure.

AI-powered invoice extraction system using a local LLM (Qwen 2.5 7B via Ollama) running entirely on-device — no data leaves the machine. The system uses pdfplumber for accurate PDF-to-text conversion with column-aware extraction, a single-pass LLM prompt to extract all invoice fields simultaneously, and a regex post-correction layer that validates and confirms extracted values. A confidence scoring system visually flags uncertain fields for human review with an amber indicator, and a Re-enforce mechanism allows users to correct extractions and teach the system vendor-specific patterns that persist for future invoices.