🤖 AI Research Agent

<div align="center">

Autonomous multi-agent AI system that generates complete research reports from natural language queries — 95% task completion rate

Overview • Architecture • Agents • Setup • Usage • Examples

</div>

---

Overview

This system lets you ask a research question in plain English and receive a structured, cited research report — automatically. It uses a swarm of specialized AI agents that collaborate, delegate tasks, and synthesize information from multiple sources.

Example:

Input:  "What are the latest advances in transformer-based protein folding models?"
Output: A 5-section research report with web sources, retrieved papers, code snippets,
        and a structured summary — generated in ~90 seconds.

Key metrics:

• ✅ 95% task success rate on 200-query benchmark
• ⚡ ~90 second average report generation time
• 📚 RAG-augmented — retrieves from a 50K-document Pinecone index
• 🔍 Real-time web search via Grok API

---

Architecture

User Query
    │
    ▼
┌─────────────────────────────────────────────────────┐
│              Orchestrator Agent (CrewAI)             │
│  - Receives query, breaks into subtasks              │
│  - Assigns tasks to specialist agents                │
│  - Synthesizes final report                          │
└──────┬───────────────┬────────────────┬──────────────┘
       │               │                │
       ▼               ▼                ▼
┌──────────────┐ ┌──────────────┐ ┌──────────────────┐
│  Web Search  │ │  RAG Agent   │ │  Code Execution  │
│    Agent     │ │              │ │     Agent        │
│              │ │ Pinecone     │ │                  │
│ Grok API     │ │ Vector Store │ │ Python REPL      │
│ (real-time   │ │ (50K docs)   │ │ (data analysis,  │
│  news/web)   │ │              │ │  calculations)   │
└──────────────┘ └──────────────┘ └──────────────────┘
       │               │                │
       └───────────────┴────────────────┘
                       │
                       ▼
              ┌─────────────────┐
              │  Report Writer  │
              │  (LangChain     │
              │   LLM chain)    │
              └────────┬────────┘
                       │
                       ▼
            Structured Research Report
            (Markdown + citations)

---

Agents

| Agent | Role | Tools | |-------|------|-------| | Orchestrator | Receives query, plans subtasks, delegates, synthesizes | CrewAI delegation | | Web Researcher | Finds current information, news, and sources | Grok API web search | | RAG Retriever | Semantic search over document knowledge base | Pinecone + embeddings | | Code Analyst | Runs Python for calculations, data analysis, plotting | Python REPL sandbox | | Report Writer | Compiles findings into structured markdown report | LangChain LLM chain |

---

Project Structure

ai-research-agent/
├── agents/
│   ├── orchestrator.py      # CrewAI orchestrator — task planning & delegation
│   ├── web_researcher.py    # Web search agent using Grok API
│   ├── rag_retriever.py     # Document retrieval agent using Pinecone
│   ├── code_analyst.py      # Python REPL execution agent
│   └── report_writer.py     # Final report synthesis agent
├── tools/
│   ├── web_search.py        # Grok API web search tool
│   ├── python_repl.py       # Sandboxed Python REPL tool
│   └── document_loader.py   # PDF/URL document loader for RAG
├── rag/
│   ├── embeddings.py        # Embedding generation (OpenAI/local)
│   ├── vector_store.py      # Pinecone index management
│   └── retriever.py         # Semantic search with re-ranking
├── config/
│   └── settings.py          # Configuration and environment variables
├── main.py                  # Entry point — run a research query
├── requirements.txt
└── README.md

---

Setup

git clone https://github.com/Kpavan3009/ai-research-agent.git
cd ai-research-agent

python -m venv venv
source venv/bin/activate

pip install -r requirements.txt

cp .env.example .env
# Fill in: GROK_API_KEY, PINECONE_API_KEY, OPENAI_API_KEY

---

Usage

# Run a research query
python main.py --query "What are the latest advances in LLM reasoning?"

# With output file
python main.py --query "Compare React vs Vue in 2025" --output reports/react_vs_vue.md

# Interactive mode
python main.py --interactive

---

Example Outputs

The agent generates structured reports like:

# Research Report: LLM Reasoning Advances (2024-2025)

## Executive Summary
...

## Key Findings
1. Chain-of-Thought prompting improvements...
2. New benchmarks showing...

## Technical Deep Dive
...

## Sources
- [Source 1](url)
- [Source 2](url)

---

Tech Stack

• Agent Framework: CrewAI, LangChain
• LLM: Grok API (xAI)
• Vector DB: Pinecone (serverless)
• Embeddings: OpenAI text-embedding-3-small
• Code Execution: RestrictedPython (sandboxed REPL)
• Document Loading: LangChain document loaders (PDF, web, YouTube)

---

<div align="center"> Made with 🔥 by <a href="https://github.com/Kpavan3009">Pavan Sesha Sai</a> </div>