PDF Insight Pro is an advanced document intelligence platform that lets you upload any PDF and interrogate its contents using natural language. Unlike conventional search tools that return raw text snippets, PDF Insight Pro runs a full agentic RAG pipeline — it chunks and embeds your document, retrieves semantically relevant passages via FAISS HNSW search, and routes the assembled context through a LangChain tool-calling agent backed by Groq-hosted LLMs. When the document alone is insufficient, the agent can reach out to Tavily’s web search API for live supplementary information, then synthesise everything into a coherent, cited answer. The result is a system that reasons about your documents rather than simply scanning them.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/Jatin-Mehra119/PDF-Insight-Beta/llms.txt
Use this file to discover all available pages before exploring further.
Key Features
PDF Ingestion & Chunking
Extracts text from any PDF using PyMuPDF, then splits it into semantically meaningful chunks with configurable size and overlap — preserving page-level metadata throughout.
FAISS HNSW Semantic Search
Builds an HNSW FAISS index over
BAAI/bge-large-en-v1.5 embeddings for sub-millisecond approximate nearest-neighbour retrieval across thousands of chunks.Agentic LangChain Pipeline
Wraps retrieval and search tools in a LangChain tool-calling agent, enabling multi-step reasoning, targeted follow-up queries, and dynamic context assembly — not just a single retrieval pass.
Multiple LLMs via Groq
Switch at request time between Llama 4 Scout 17B, Llama 3.1 8B, Llama 3.3 70B, and GPT-OSS 120B — all served at high throughput through the Groq inference API.
Tavily Web-Search Augmentation
Optionally extend every answer with live web results via the Tavily Search API — useful for documents that reference time-sensitive data, external standards, or evolving topics.
Session Persistence
Every upload creates an isolated session that persists the FAISS index, chunk store, and full chat history to disk, so you can continue a conversation across requests without re-uploading.
Tech Stack
| Layer | Technology |
|---|---|
| Backend | Python 3.12, FastAPI, Uvicorn, PyMuPDF, LangChain, FAISS, Sentence Transformers, Groq API, Tavily API |
| Frontend | HTML / CSS / JavaScript, Font Awesome, Highlight.js, Marked.js |
| DevOps | Docker, GitHub Actions CI/CD, Hugging Face Spaces |
RAG Evaluation Metrics
PDF Insight Pro was benchmarked against theneural-bridge RAG evaluation dataset. The table below summarises the key performance indicators.
| Metric | Value |
|---|---|
| Semantic Similarity (Mean) | 0.852 |
| ROUGE-L F1 Score (Mean) | 0.395 |
| Semantic Similarity (Max) | 1.000 |
| ROUGE-L F1 Score (Max) | 1.000 |
| Semantic Similarity (Min) | 0.592 |
| ROUGE-L F1 Score (Min) | 0.099 |
| Standard Deviation (Similarity) | 0.089 |
| Standard Deviation (ROUGE-L F1) | 0.217 |
| Status | Count | Percentage |
|---|---|---|
| PASS | 64 | 85.3 % |
| FAIL | 11 | 14.7 % |
PDF Insight Pro is released under the MIT License, meaning you are free to use, modify, and distribute it in personal and commercial projects. A live demo is available on Hugging Face Spaces — no local setup required to try it out.