🧠 Agentic AI Application — Multi-Agent Supervisor System

A full-stack, modular, agentic AI system powered by LangGraph, LangChain, and multiple external tools. Built to handle intelligent task decomposition, multi-step reasoning, and user-friendly interaction.

📸 Preview

Chatbot UI (1)	Chatbot UI (2)

User Auth	Workflow

FastAPI	LangSmith

---

🧩 Architecture Overview

                      ┌────────────┐
                      │ Supervisor │
                      └─────┬──────┘
                            ↓
        ┌──────────┬──────────────┬─────────────┬───────────────┐
        ↓          ↓              ↓             ↓               ↓
    Greeting    Enhancer       Coder     Maths Reasoner     Researcher
        │          │              │             │               │
        └──────────┴────┬─────────┴─────────────┴───────────────┘
                        ↓
                 Tool Execution Layer
            (Python REPL, Tavily, PubMed, etc.)

---

🛠️ Tech Stack

Layer	Tools/Frameworks
Agents & Logic	LangChain, LangGraph
Backend	FastAPI, Supabase (Auth & DB)
Frontend	Streamlit with real-time chat streaming
DevOps	Docker, GitHub, Langsmith

---

🧠 Agents & Tools

👷 Supervisor Agent

• Central brain of the system

• Delegates tasks to other agents

• Makes decisions based on outputs and history

✨ Enhancer Agent

• Improves clarity, tone, and relevance of intermediate results

💬 Greeting Agent

• Handles user onboarding and initial context setting

🧑‍💻 Coder Agent

• Writes and debugs code

• Uses PythonREPL for live execution

🧠 Maths Reasoner

• Handles symbolic & numerical problems

• Uses calculator and PythonREPL

🔍 Researcher

• Finds factual data and references using:

  • Tavily

  • DuckDuckGo

  • Wikipedia

  • PubMed

---

💡 Features

✅ Real-time streaming chatbot interface
✅ Multi-session / Multi-chat support
✅ User authentication via Supabase
✅ Modular agent-tool orchestration
✅ LangSmith monitoring and observability
✅ Dynamic memory per chat
✅ Fully containerized with Docker
✅ Ready for deployment on Render etc.

---

🧰 Setup Instructions

1. Clone Repo

git clone https://github.com/ayushWeb07/supervisor-multi-agent-app.git
cd supervisor-multi-agent-app

2. Environment Setup

Create .env file:

TAVILY_API_KEY=...
LANGSMITH_TRACING=...
LANGSMITH_ENDPOINT=...
LANGSMITH_API_KEY=...
LANGSMITH_PROJECT=...
OPENAI_API_KEY=...
GROQ_API_KEY=...
SUPABASE_URL=...
SUPABASE_KEY=...

3. Run Locally

# Backend
uvicorn backend.fastapi_backend:app --reload

# Frontend
streamlit run .\frontend\streamlit_frontend.py

---

🧪 Tooling Integration

Tool	Purpose
Tavily	Fast multi-source web search
Wikipedia	Structured encyclopedic info
DuckDuckGo	Lightweight search fallback
PubMed	Scientific paper lookup
PythonREPL	Code execution sandbox
Calculator	Basic arithmetic + formulas

---

🔗 Links:

🎥 Full App Demo | 📖 Medium Blog | 👨‍💼 LinkedIn Post

🤝 Contributing

PRs welcome! If you're into agentic systems, AI agents, and RAG pipelines, feel free to fork, tinker, and improve.