🧠 Research Intelligence Agent

Multi-Agent Research Intelligence System | Agents League Hackathon 2026

---

🎯 Problem Statement

Information on the web is fragmented, contradictory, and difficult to trust. Researchers, students, and professionals spend hours manually searching, reading, cross-referencing, and synthesising sources — only to still be uncertain about what to believe.

Research Intelligence Agent solves this by deploying a team of 5 specialised AI agents that autonomously search, read, fact-check, and synthesise information into a structured, confidence-scored research report — in seconds.

---

🏗️ Architecture

User (Browser)
     │
     ▼
React Frontend  ──────────────────  Azure Static Web Apps (free tier)
     │ HTTP/REST
     ▼
FastAPI Backend  ─────────────────  Azure App Service
     │
     ▼
┌─────────────────────────────────────────┐
│       ORCHESTRATOR AGENT                │ ◄── Foundry IQ (Microsoft IQ Layer)
└──────┬──────────┬──────────┬────────────┘
       ▼          ▼          ▼
  Search      Reader     Fact-Check ──┐
  Agent       Agent        Agent      │ open gaps found?
       ▲                    │         │
       └────────────────────┴─────────┘  yes → targeted follow-up search/read, re-check
                            │ no
                            ▼
                     Synthesis Agent
                            ▼
     Confidence-Scored Research Report

Agent Roles

Agent	Responsibility
Orchestrator	Runs the multi-step reasoning loop — decides whether to search again based on what Fact-Check finds, and manages agent handoffs via Foundry IQ
Search Agent	Uses Tavily Search API to find real-time web sources
Reader Agent	Fetches and extracts key facts from web pages
Fact-Check Agent	Cross-references sources, detects contradictions, and flags evidence gaps
Synthesis Agent	Produces final report with confidence-scored findings

🧭 Multi-Step Reasoning Loop

The orchestrator doesn't just run the pipeline once and stop. After the Fact-Check Agent reports its findings, the orchestrator inspects the Gaps section: if it identifies an open question the initial sources didn't answer, the orchestrator formulates a targeted follow-up query, sends the Search and Reader Agents out again, merges the new evidence with the old (de-duplicated by URL), and re-runs the Fact-Check Agent — before finally handing everything to Synthesis. This loop is capped by MAX_RESEARCH_ROUNDS (default 2) so the agent always converges on a report, while still demonstrating genuine "notice what I don't know → go find it" reasoning rather than a fixed, one-shot pipeline.

---

💡 Microsoft Technologies Used

Technology	Role
Azure AI Foundry	Powers all 5 agents via Foundry IQ intelligence layer
Tavily Search API	Real-time web search for the Search Agent
Azure App Service	Hosts the FastAPI Python backend
Azure Static Web Apps	Hosts the React frontend
Azure Managed Identity	Passwordless authentication (DefaultAzureCredential)
Azure API Key auth	Fallback for hosted environments without Managed Identity

---

🚀 Local Setup — Step by Step

Prerequisites

• Python 3.10+
• Node.js 18+
• Azure account (free at portal.azure.com)
• Azure CLI installed and logged in (az login)
• Azure AI Foundry project created with:
  • gpt-4o model deployed
  • Grounding with Bing Search resource connected

Step 1 — Unzip and enter folder

unzip research-intelligence-agent.zip
cd research-intelligence-agent

Step 2 — Python virtual environment

# Mac/Linux
python -m venv .venv
source .venv/bin/activate

# Windows
python -m venv .venv
.venv\Scripts\activate

Step 3 — Install Python dependencies

pip install -r requirements.txt

Step 4 — Install frontend dependencies

cd frontend
npm install
cd ..

Step 5 — Configure credentials

cp .env.example .env
# Open .env in a text editor and fill in your values

Your .env must contain:

AZURE_AI_FOUNDRY_PROJECT_ENDPOINT=https://YOUR-RESOURCE.services.ai.azure.com/api/projects/YOUR-PROJECT
AZURE_API_KEY=your-azure-api-key
AZURE_AI_MODEL_DEPLOYMENT=gpt-4o
TAVILY_API_KEY=your-tavily-api-key

Step 6 — Login to Azure

az login

Step 7 — Run the app

# Mac/Linux
./start.sh

# Windows
start.bat

Step 8 — Open in browser

Go to http://localhost:5173

---

📁 Project Structure

research-intelligence-agent/
├── agents/
│   ├── orchestrator.py      # Master coordinator (Foundry IQ entry point)
│   ├── search_agent.py      # Source discovery via Grounding with Bing
│   ├── reader_agent.py      # Content extraction from web pages
│   ├── factcheck_agent.py   # Cross-referencing & contradiction detection
│   ├── synthesis_agent.py   # Final confidence-scored report generation
│   └── base_agent.py        # Azure AI Foundry base class (all agents)
├── tools/
│   ├── web_search.py        # Grounding with Bing Search integration
│   └── document_reader.py   # Web page content extractor
├── frontend/                # React web UI
│   ├── src/
│   │   ├── App.jsx          # Main UI with live agent progress
│   │   └── App.css          # Microsoft-themed dark UI
│   ├── index.html
│   └── package.json
├── outputs/                 # Generated reports saved here
├── app.py                   # FastAPI backend entry point
├── config.py                # Environment configuration
├── requirements.txt         # Python dependencies
├── .env.example             # Credentials template
├── start.sh                 # One-command startup (Mac/Linux)
├── start.bat                # One-command startup (Windows)
└── README.md

---

🔒 Safety & Reliability

• No confidential data processed or stored
• Input validation on all user queries
• Graceful error handling at every agent step — fallbacks if a source fails
• No hardcoded credentials — all via environment variables
• Managed Identity used for Azure authentication (no API keys stored)
• Confidence scores make uncertainty explicit — never presents guesses as facts
• Web content treated as untrusted input — validated before agent processing

---

🎥 Demo Video

Link : https://research-intelligence-agent.onrender.com/

---

👤 Author

Karthick Subramanian || Esvanth Mohankumar | MSc Computing (AI), National College of Ireland,Dublin,Ireland
Agents League Hackathon 2026 | Reasoning Agents Track | Foundry IQ