README.md

metadata

title: Agentic Document Intelligence
emoji: 📄
colorFrom: blue
colorTo: pink
sdk: gradio
sdk_version: 6.2.0
app_file: app.py
pinned: false
license: apache-2.0

📄 Agentic Document Intelligence

PDF RAG with Together.ai

This Hugging Face Space demonstrates a Retrieval-Augmented Generation (RAG) system that allows users to upload a PDF and ask questions that are strictly grounded in the document content.

The Space serves as a foundational Agentic Document Intelligence component, designed to be simple, transparent, and extensible.

---

🚀 What This Space Does

• Upload a PDF document
• Build a semantic index using embeddings + FAISS
• Ask natural-language questions
• Receive answers grounded only in the uploaded document
• View retrieved source passages for transparency

---

🧠 Architecture Overview

1. PDF Ingestion

   • Extracts text from uploaded PDF
   • Cleans and normalizes content

2. Chunking

   • Splits text into overlapping semantic chunks
   • Ensures contextual continuity

3. Vector Indexing

   • Generates embeddings using Sentence Transformers
   • Indexes vectors using FAISS (cosine similarity)

4. Retrieval

   • Retrieves top-K relevant chunks for each query

5. Generation (RAG)

   • Injects retrieved context into LLM prompt
   • Uses Together.ai (Mixtral) for answer generation

---

▶️ How to Use This Space (End-to-End)

Step 1: Upload a PDF

• Click “Upload PDF”
• Select a text-based PDF file

  ⚠️ Note: Scanned PDFs without text extraction will not work unless OCR is applied.

---

Step 2: Wait for Indexing

• The system will:
  • extract text
  • split it into chunks
  • build a FAISS vector index
• You will see a confirmation message:

---

Step 3: Ask a Question

• Type a natural-language question related to the document
  Examples:
• “Summarize the document”
• “What is the main contribution?”
• “Explain the methodology section”

---

Step 4: Receive the Answer

You will get:

• ✅ A generated answer based only on document context
• 📌 Retrieved source passages with similarity scores
• 🚫 No hallucinated or external information

If the answer is not present in the document, the system will respond:

---

Step 3: Ask a Question

• Type a natural-language question related to the document
  Examples:
• “Summarize the document”
• “What is the main contribution?”
• “Explain the methodology section”

---

Step 4: Receive the Answer

You will get:

• ✅ A generated answer based only on document context
• 📌 Retrieved source passages with similarity scores
• 🚫 No hallucinated or external information

If the answer is not present in the document, the system will respond:

---

🤖 Models Used

Language Model

• Provider: Together.ai
• Model: mistralai/Mixtral-8x7B-Instruct-v0.1

Embedding Model

• sentence-transformers/all-MiniLM-L6-v2

---

🧰 Tech Stack

• Python
• Gradio (UI)
• FAISS (vector search)
• Sentence Transformers (embeddings)
• Together.ai (LLM)
• Hugging Face Spaces

---

🔐 Environment Configuration (For Developers)

Secrets

• TOGETHER_API_KEY → Together.ai API key
• OPENAI_API_KEY → Same value (compatibility with OpenAI client)

Variables

• TOGETHER_MODEL → mistralai/Mixtral-8x7B-Instruct-v0.1
• TOGETHER_BASE_URL → https://api.together.xyz/v1

---

🧩 Intended Use Cases

• Research paper Q&A
• Technical documentation assistants
• Internal knowledge bases
• RAG pipeline reference implementation
• Agentic AI system foundations

---

🔮 Future Enhancements

• Multi-PDF support
• Chat memory
• Streaming responses
• Agent routing & tool usage
• Evaluation and scoring agents

---

🙌 Author

Built by Abhishek Prithvi Teja
Focused on Agentic AI, RAG systems, and applied LLM engineering

---

🏷️ Tags

rag · agentic-ai · document-qa · faiss · together-ai · huggingface-spaces