🚀 Gen-AI Playground (LangChain + Azure OpenAI)

This repository documents my hands-on exploration of LLM systems, tool calling, and agent design using LangChain.

Instead of relying on high-level abstractions, this repo focuses on understanding how things work under the hood — building components manually and then layering applications on top.

---

🧠 What this repo covers

• 🔧 Manual LLM tool-calling orchestration (agent internals)
• 🌐 Integration with external tools (Wikipedia, Tavily Search)
• 💬 Stateful conversational apps using Streamlit
• ⚙️ Azure OpenAI integration via LangChain
• 🧩 Prompt pipelines using LCEL (prompt | model | parser)
• 🧪 Debuggable message-driven workflows

---

📁 Project Structure

Gen-AI/
│
├── Langchain/
│   ├── QA-ChatBot.py
│   └── manual_tool_calling_agent.ipynb
│
├── main.py
├── pyproject.toml
├── requirements.txt
├── .gitignore
└── README.md

---

🤖 Streamlit QA Chatbot

A simple conversational chatbot built using:

• Streamlit UI
• Azure OpenAI (GPT-4o / GPT-4.1-mini)
• LangChain LCEL pipeline

✨ Features

• Model selection from UI
• Streaming responses (token-by-token)
• Persistent conversation using session_state
• Prompt → Model → Parser pipeline
• Environment-based configuration via .env

⚙️ Flow

User Input → Prompt Template → AzureChatOpenAI → Output Parser → UI

---

🧠 Manual Tool-Calling Agent (Core Highlight)

File: manual_tool_calling_agent.ipynb

This is the core learning component where the full agent loop is implemented manually:

1. User query sent to LLM
2. LLM returns tool_calls
3. Tools executed manually in Python
4. Results returned via ToolMessage
5. Loop continues until final answer

🔁 Flow

User → LLM → Tool Calls → Tool Execution → LLM → Final Answer

🔧 Capabilities

• Custom tools (add, multiply)
• External tools (Wikipedia, Tavily)
• Dynamic tool routing
• Multi-step reasoning
• Error handling
• Full message trace for debugging

---

⚙️ Tech Stack

• Python 3.12
• LangChain
• Azure OpenAI
• Streamlit
• Tavily API
• Wikipedia API
• uv (package manager)

---

🛠️ Setup Instructions

1. Clone the repository

git clone https://github.com/BrainTeaser1/Gen-AI.git
cd Gen-AI

2. Install dependencies (using uv)

uv sync

or

uv pip install -r requirements.txt

3. Create .env

AZURE_OPENAI_API_KEY=your_key_here
AZURE_OPENAI_ENDPOINT=your_endpoint_here
TAVILY_API_KEY=your_key_here

4. Run the Streamlit App

streamlit run Langchain/QA-ChatBot.py

---

🔐 Security Note

• .env is ignored via .gitignore
• No secrets are stored in the repository
• API keys must be configured locally

---

🚧 Current Status

• ✅ Manual tool-calling agent implemented
• ✅ Multi-tool reasoning loop working
• ✅ Streamlit chatbot UI built
• ✅ Azure OpenAI integration complete

---

🔭 What's Next

• LangGraph-based orchestration
• Memory (short + long term)
• RAG integration
• Observability (logging, token tracking)
• Tool reliability improvements

---

👤 Author

Krishna Shukla

Exploring:

• LLM Systems
• Agent Architectures
• GenAI Infrastructure

---

⭐ Final Note

This repository focuses on understanding before abstraction.

Instead of relying on built-in agents, I implemented the core loop manually — making it easier to debug, extend, and reason about system behavior.