Advanced Document Intelligence Platform
Powered by Vision-Language Models and Retrieval-Augmented Generation
📄
OCR Endpoint
Extract text from historical PDFs using Llama-4-Maverick-17B Vision model
- Multi-language support (Azerbaijani, Russian, English)
- Handwriting recognition
- Image detection and referencing
- 88.3% Character Success Rate
🤖
LLM Endpoint
Ask questions about historical documents with RAG-powered chatbot
- Retrieval-Augmented Generation (RAG)
- 1,128 vectors from 28 documents
- Citation-focused responses
- Top-3 document retrieval
Technical Stack
OCR Model
Llama-4-Maverick-17B
Embedding Model
BAAI/bge-large-en-v1.5
Vector Database
Pinecone (1024 dims)
LLM Model
Llama-4-Maverick-17B
Framework
FastAPI + Docker
Documents
28 PDFs, 1,128 vectors
88.3%
OCR Accuracy (CSR)
1,128
Total Vectors
28
Documents Indexed
~2.6s
Avg Response Time