Hybrid Search

Combines dense embeddings (BAAI/bge-small-en-v1.5) with sparse BM25 vectors for superior semantic search results. Optional cross-encoder reranking for enhanced precision.

Document Processing

Async pipeline supporting PDF, DOCX, TXT, MD, and image formats. RapidOCR extracts text from uploaded images and low-text PDF pages automatically.

Document Classification

Local ModernBERT ONNX classification for automatic categorization. Ships with a built-in financial and corporate document taxonomy, fully configurable via environment variables.

Entity Extraction

Built-in spaCy NER extracts persons, organizations, dates, locations, monetary amounts, account numbers, and transaction references. Entities are stored alongside embeddings for rich metadata filtering.

Async Processing

Celery-powered background processing with dedicated preprocessing and embedding queues. Scales horizontally by adding more workers where you need them.

Production Ready

Dockerized deployment with health checks, memory management, structured logging, and live status dashboards. Built for real-world enterprise workloads.