🧠 AI Workflow Assistant (ReAct Framework + LangGraph + Groq)

An AI-powered workflow automation assistant that can plan, reason, and act using tools to analyze raw CSV data and generate structured analytical reports.

This project demonstrates true ReAct-style reasoning, LangGraph agent loops, tool calling, and Groq LLM integration — making it a perfect portfolio-ready example of building real-world AI systems.

---

🚀 Features

🔹 ReAct (Reason + Act) Agent

Uses LangGraph’s state machine to plan → call tools → observe results → think → act again… until a final answer is produced.

🔹 Tool-Calling Capabilities

Includes 3 built-in data tools:

• load_csv_overview — summarize dataset
• describe_column — statistics for numeric columns
• filter_rows — Pandas-powered conditional filtering

Agents intelligently choose when to call tools.

🔹 Structured Final Report

The agent outputs a Markdown report with sections:

1. Overview of Data
2. Key Metrics & Trends
3. Detailed Insights
4. Anomalies / Data Quality Notes
5. Recommendations

🔹 Groq LLM Integration

Fast & cost-efficient inference using ChatGroq (e.g., Llama-3.3-70B, Llama-3.1, Gemma 2, etc.)

---

📂 Project Structure

ai-workflow-assistant/
├── .env
├── requirements.txt
├── data/
│ └── sample_data.csv
└── src/
├── init.py
├── config.py
├── tools.py
├── state.py
├── graph.py
└── main.py

▶️ Usage

1. Place your CSV file in the data/ folder or use an absolute path.
2. Run the agent:

python -m src.main --data sample_data.csv --task "Analyze regional sales performance."

3. Example output:

-------------------------------
     Running AI Workflow Assistant
-------------------------------

Final Report:
# Sales Analysis Report
...

🧩 How It Works (Architecture)

1. This project implements the ReAct Framework using LangGraph.

• Agent Node
• Receives messages
• Calls the Groq LLM
• Determines if a tool should be used

2. 🔸 Tool Node

• Executes actual Python functions (Pandas-backed tools).

3. 🔸 Conditional Edge

should_continue decides:

• If tool calls exist → run tools
• Otherwise → stop and produce final answer

This creates a loop:

Think → Act (Tool) → Observe → Think → ... → Final Answer

🛠️ Tech Stack

• LangGraph – agent orchestration & state machine
• LangChain – messages, tools, LLM interface
• Groq API – ultra-fast inference with Llama models
• Python
• Pandas – data operations
• Rich – nice terminal output

📈 Future Enhancements

• Add planning node (separate plan → execute graph)
• Support Excel, Parquet, SQL sources
• Add more data analysis tools (outliers, trends, clustering)
• Web UI using Streamlit or FastAPI
• Export final report to Markdown/PDF

Author

Created by Syed Waleed Ahmed