🚀 GraphLens AI

GraphLens AI is an AI-powered research workspace. It integrates document-based vector search (RAG), real-time web search, and an autonomous file management assistant into a secure, user-isolated chat platform.

---

📸 Demo / Preview

---

🏗️ Interactive System Architecture

🚀 Click here to open the Live Interactive Architecture Diagram

---

🔄 System Flow

1. Ingestion & Storage: The user uploads a PDF or text document. The Django backend parses text using PyMuPDF, persists the file to AWS S3 or local storage, and logs the metadata in the relational database (UserFile).
2. Vector Indexing: Extracted content is chunked and embedded using all-MiniLM-L6-v2 or llama-text-embed-v2, then indexed into Pinecone Serverless or local ChromaDB, tagged with the user's user_id for multi-tenant isolation.
3. User Query: The user sends a prompt through the chat interface, optionally targeting active documents.
4. Supervisory Routing: The LangGraph Supervisor inspects the query using fast deterministic checks and structured LLM routing to delegate tasks across 4 specialized worker agents.
5. Agent Execution:
   • 📄 Document RAG Agent: Performs similarity searches across the vector database (Pinecone/ChromaDB) scoped to the user's active document and retrieves top context chunks.
   • 🔍 Web Researcher Agent: Conducts live internet searches on DuckDuckGo and deep-scrapes markdown text using Jina Reader (r.jina.ai) in an autonomous ReAct loop.
   • 📁 Workspace File Agent: Creates, reads, updates, or deletes files directly inside the user's workspace on disk/S3, automatically triggering vector store re-indexing.
   • 📧 Email Agent: Extracts recipient and document parameters, resolves attachments from storage/UserFile database, and dispatches emails via SMTP relay with custom Reply-To headers.
6. Synthesis & Response: Context is synthesized by the LLM (Groq Cloud Llama-3.3-70B or Local Ollama Llama-3.1) into a grounded answer, returned via Django REST Framework, and rendered live in the chat dashboard.

---

🚀 Key Features

🧠 Intelligent Agent Routing (LangGraph)

• Built using LangGraph Supervisor Architecture.
• Intent-Based Dynamic Routing: Automatically delegates queries between document RAG, web search, file management, and email dispatch.

🔍 Isolated Document RAG (Dual Vector Store)

• Multi-Tenant Privacy: Scopes document ingestion and vector retrieval by user session to ensure complete data isolation.
• Flexible Vector Backend: Seamlessly toggle between Pinecone Serverless (cloud) and ChromaDB (local disk).
• Semantic Embeddings: Powered by HuggingFace all-MiniLM-L6-v2 or Pinecone llama-text-embed-v2.

🌐 Real-Time Web Research

• Live Search: Fetches real-time internet facts using DuckDuckGo.
• URL Deep-Scraping: Converts web pages into clean markdown via the Jina Reader API for the LLM to inspect.

📁 Chat-Based File Manager & Auto-Sync

• Workspace CRUD: Manage server and cloud files (create, read, append, overwrite, delete) directly through chat instructions.
• Auto-Sync Indexing: Background workers automatically re-embed and index modified files so vector search stays current.

📧 Automated Email Dispatch Agent

• Attachment Delivery: Sends workspace documents, PDFs, and generated reports directly to any recipient email.
• Sender Identity: Dispatches via central SMTP relay while attaching user identification and setting Reply-To headers.

💻 Hybrid AI Engine (Cloud + Local Offline)

• Groq Cloud Engine: High-speed inference using llama-3.3-70b-versatile.
• Local Offline Engine: 100% private, local inference using Ollama (llama3.1).

---

🎯 Use Cases

• 📚 Academic & Enterprise Research: Ask questions about reference PDFs and verify claims with live web searches.
• 📄 Workspace Note-Taking & Drafting: Summarize documents, draft reports, and save them directly as workspace files.
• ✉️ Direct Document Sharing: Tell the agent to "send the summary report to colleague@example.com" and have it dispatched with attachment.
• 🔒 Privacy-First Offline Analysis: Switch to Ollama + ChromaDB to run the entire RAG pipeline completely offline.

---

🛠️ Tech Stack

• Frontend: Vanilla HTML5, CSS3 (Glassmorphism theme), and JavaScript.
• Backend: Django & Django REST Framework (DRF).
• AI Orchestration: LangChain & LangGraph (Supervisor multi-agent state graph).
• AI Inference (Hybrid):
  • Cloud: Groq Cloud API (llama-3.3-70b-versatile).
  • Local / Offline: Ollama (llama3.1).
• Embeddings: HuggingFace (all-MiniLM-L6-v2) & Pinecone Embeddings (llama-text-embed-v2).
• Vector Databases: Pinecone Serverless (cloud) & ChromaDB (local persistent).
• Relational Database: PostgreSQL / SQLite (Django ORM with UserFile & auth models).
• Storage Vault: AWS S3 (via boto3 presigned URLs) & local user media cache.
• Email & External APIs: SMTP Mail Relay, DuckDuckGo Search, and Jina Reader API.

---

🚀 Future Roadmap

• 🔹 Google Workspace Integration (Docs, Drive, Gmail)
• 🔹 Multi-File Chat: Simultaneous querying across multiple documents.
• 🔹 Persistent Memory: Longer-term session history for returning users.
• 🔹 Enterprise Auth: Production-grade user authentication systems.

---

📬 Contact

📧 vinaygattu005@gmail.com
🔗 GitHub: Vinay50029

---

⭐ If you like this project, give it a star on GitHub — it helps a lot!