PDFhelper: Intelligent Document Retrieval
RAG-based system for efficient document querying and management.
LLaMA 3.2QdrantHugging Face
Problem
Managing and extracting insights from large collections of PDF documents is a bottleneck for research. Standard keyword search often misses semantic context.
Solution: RAG Pipeline
PDFhelper is a Retrieval-Augmented Generation (RAG) system that uses vector embeddings to understand the "meaning" of queries.
Tech Stack
- Embeddings: Hugging Face
all-MiniLM-L6-v2. - Vector DB: Qdrant (local Docker instance).
- LLM: LLaMA 3.2 (via Ollama) for generation.
- Orchestration: LangChain.
Workflow
- Ingestion: PDFs are parsed, chunked (500 chars with 50 overlap), and embedded.
- Storage: Vectors stored in Qdrant collections.
- Retrieval: User query is embedded; Top-K cosine similarity match found.
- Generation: Retrieved chunks + Query sent to LLaMA 3.2 to generate the answer.
Impact
- Reduced search time by ~60% for technical documentation.
- Enable "Chat with your Data" functionality locally without data leaving the machine.