🚀 DeepSight Engine

A production-grade Agentic Deep Research Engine built with LangGraph that autonomously plans, researches, critiques, and generates citation-backed research reports using multiple collaborating AI agents.

---

🏗️ System Architecture

The architecture below illustrates the complete end-to-end workflow of DeepSight Engine.

---

---

📸 Application Preview

🏠 Home Screen

---

🔄 Research Progress

Real-time streaming of agent execution using Server-Sent Events (SSE).

---

📑 Generated Research Report

Structured, citation-backed report generated after the multi-agent workflow completes.

---

✨ Key Features

• 🧠 Dynamic query decomposition using a dedicated Planner Agent
• 🔎 Autonomous multi-step web research with Researcher Agent
• 📝 Structured report generation with Reporter Agent
• 🧐 Quality evaluation through a Critic Agent
• ✍️ Automatic refinement using an Editor Agent
• 🌐 Multi-provider LLM Gateway (Llama, OpenAI & Claude)
• ⚡ Real-time progress streaming via Server-Sent Events (SSE)
• 🧠 Redis Semantic Caching for repeated and semantically similar queries
• 📄 PDF & DOCX report export
• 📊 Usage tracking and API key management
• 🚀 Production-ready FastAPI backend

---

🤖 Multi-Agent Workflow

Unlike a traditional chatbot that relies on a single LLM response, DeepSight Engine orchestrates multiple specialized AI agents.

🧠 Planner Agent

• Understands the user's objective
• Breaks complex queries into dynamic research objectives
• Creates an execution strategy

🔎 Researcher Agent

• Executes one research objective at a time
• Performs iterative web research
• Collects and consolidates evidence

📝 Reporter Agent

• Synthesizes collected information
• Generates a structured research report

🧐 Critic Agent

• Evaluates report quality
• Detects missing information
• Decides whether refinement is required

✍️ Editor Agent

• Improves weak sections
• Refines report quality
• Produces the final polished output

---

🔀 LLM Gateway

DeepSight includes an intelligent LLM Gateway supporting multiple providers.

Supported Providers

• Llama
• OpenAI
• Claude

Capabilities

• Smart model routing
• API key management
• Usage tracking
• Provider abstraction

---

⚡ Technology Stack

Backend

• FastAPI
• Python
• LangGraph
• LangChain

AI

• OpenAI
• Claude
• Llama
• Tavily Search API

Databases

• PostgreSQL
• Redis

Infrastructure

• Server-Sent Events (SSE)
• Docker
• Render

---

🔄 Workflow

User Query
      │
      ▼
Planner Agent
      │
      ▼
Dynamic Research Plan
      │
      ▼
Researcher Agent
      │
      ▼
Web Search (Tavily)
      │
      ▼
Reporter Agent
      │
      ▼
Critic Agent
      │
      ▼
Editor Agent
      │
      ▼
Citation-backed Research Report

---

🚀 Why DeepSight?

Traditional AI chatbots usually perform a single LLM call.

DeepSight instead follows a production-style multi-agent workflow where specialized agents collaborate to:

• Plan
• Research
• Verify
• Critique
• Refine

This produces more comprehensive, citation-backed reports while keeping the architecture modular and extensible.

---

⚙️ Running Locally

Clone the repository

git clone https://github.com/<your-username>/DeepSight-Engine.git

Navigate into the project

cd DeepSight-Engine

Install dependencies

pip install -r requirements.txt

Create a .env file

LLAMA_API_KEY=
OPENAI_API_KEY=
CLAUDE_API_KEY=
TAVILY_API_KEY=
DATABASE_URL=
REDIS_URL=

Run the application

uvicorn app.main:app --reload

Open

http://localhost:8000

---

👨‍💻 Author

Oishik Bandyopadhyay

AI Engineer • GenAI • Agentic AI • LangGraph • LLM Engineering

---

⭐ If you found this project interesting, consider giving it a star!