Augmented-Research-Agent-with-LLM-Tooling

🔍 Augmented Deep Research Agent

LLM-Powered Autonomous Research System

This project is a production-aligned Deep Research Agent that performs fully automated research using:

• Google Gemini LLM
• Google Search API
• Autonomous task decomposition
• Evidence aggregation
• Contradiction detection
• Final structured research synthesis
• FastAPI backend
• Simple dark-theme web UI

Designed for real-world GenAI applications and agentic systems.

---

⭐ Features

🔹 Intelligent Multi-Step Research

The agent:

• Breaks research questions into sub-tasks
• Decides which tasks require web search
• Retrieves information via Google Search
• Synthesizes findings using Gemini LLM
• Removes duplicates and low-quality results

🔹 Structured, Professional Output

Returns a JSON research report with:

• sections: Summaries for each sub-task
• conclusion: Final combined insight
• citations: Clean URL references
• contradictions_and_uncertainties: If detected
• Always deterministic and clean

🔹 FastAPI Backend

• /research → POST research queries
• / → Serves minimal UI
• /docs → Interactive Swagger documentation

🔹 Simple Modern UI (Dark Mode)

• Clean, non-chatbot interface
• Textbox → button → formatted research output
• Clickable citations

---

🧠 Architecture

User Query ↓ TaskDecomposer (Gemini LLM) ↓ Research Controller ├── LLM (Gemini) └── Search Tool (Google Search API) ↓ Synthesizer (LLM backed) ↓ Structured Research JSON

---

📂 Project Structure

Augmented Research Agent with LLM Tooling/ │ ├── agent/ │ ├── controller.py │ ├── task_decomposition.py │ ├── synthesizer.py │ ├── models/ │ ├── base.py │ ├── gemini.py │ ├── tools/ │ ├── base.py │ ├── search.py │ ├── static/ │ ├── index.html # Dark minimal UI │ ├── api.py # FastAPI server ├── main.py # CLI mode ├── requirements.txt ├── .gitignore └── README.md

🛠️ Tech Stack

Layer / Purpose	Technology Used	Description
LLM Engine	Google Gemini (1.5 Flash / Pro)	Generates summaries, decomposes tasks, synthesizes research
Search Engine	Google Custom Search API	Retrieves real-time web data for research
API Backend	FastAPI	High-performance web framework for research endpoints
Server Runner	Uvicorn (ASGI)	Fast server for hosting FastAPI app
UI Framework	TailwindCSS + Vanilla JS	Simple, clean, dark-mode front-end
HTTP Requests	Requests (Python)	Used for Google Search API calls
Environment Mgmt	python-dotenv	Secure loading of API keys from .env
Data Validation	Pydantic	Validates request/response models in FastAPI
Typing Support	typing_extensions	Ensures strong typing for Python < 3.12
Runtime Environment	Python 3.x	Core programming language