🧠 Agentic AI Chat Analyzer

An AI-powered platform for analyzing agent chat transcripts about Washington Post articles using lightweight LLMs, visual EDA, and an interactive frontend.
Designed to summarize conversations, detect sentiments, and deliver insights — all through a modular, API-driven architecture.

---

🚀✨ Features

✅ Modular Data Pipeline – Clean ingestion, cleaning, and transformation

📊 EDA + Profiling – Dataset summaries, word clouds, sentiment plots

🧠 LLM Summarizer – Uses Flan-T5-small

💬 Sentiment Classification – CardiffNLP RoBERTa Sentiment

🌐 FastAPI Backend – 3 REST endpoints with Pydantic validation

🖼️ Streamlit Frontend – Interactive UI for real-time transcript analysis

📦 Model Caching – Offline-ready with locally saved models

🧪 Evaluation Pipeline – Accuracy and BLEU score metrics

---

🗂️ Folder Structure

Agentic_AI_ChatAnalyzer/
├── app/                # Core app logic (API, services, utils)
├── data/               # Dataset and output CSVs
├── models/             # Locally saved HuggingFace models
├── notebooks/          # EDA and profiling notebook
├── scripts/            # Model downloader and batch inference
├── streamlit_app/      # UI client (integrated with FastAPI)
├── main.py             # FastAPI entrypoint
├── requirements.txt    # Python dependencies
├── .gitignore
└── README.md

---

📐 System Architecture

flowchart LR
    A["User Input via Streamlit UI"] --> B["POST Transcript to FastAPI API"]
    B --> C["LLM Summarizer (Flan-T5)"]
    B --> D["Sentiment Classifier (RoBERTa)"]
    C --> E["Summary Output"]
    D --> F["Sentiment per Agent"]
    E --> G["Response JSON to Streamlit"]
    F --> G
    G --> H["Visualization: WordClouds + Metrics"]

Loading

---

⚙️ Setup Instructions

# 1. Clone the repo
git clone https://github.com/yashdew3/Agentic-AI-Chat-Analyzer.git
cd Agentic_AI_ChatAnalyzer

# 2. Create and activate environment (optional)
python -m venv venv
source venv/bin/activate   # Windows: venv\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Download local LLM + sentiment models (run once)
python scripts/download_models.py

---

🚀 How to Use

Run FastAPI Backend

uvicorn main:app --reload

• Access Swagger Docs: http://localhost:8000/docs

---

Run Streamlit Frontend

streamlit run streamlit_app/app_ui.py

• Analyze transcripts visually

• Paste chats and see live:

  • Summary
  • Sentiment Insights
  • Word Clouds
  • Article Links

---

📡 API Endpoints

Method	Endpoint	Description
GET	/summary	Returns dataset stats
POST	/transform	Preprocess and clean new chat input
POST	/insights	Summarizes, classifies sentiment, returns insights

---

🧪 Sample API Input

📥 Sample Payload (for /insights)

{
  "content": [
    {"agent": "agent_1", "message": "Let’s discuss the article on the new football rule change."},
    {"agent": "agent_2", "message": "Yes, it's causing a lot of debate on ESPN."}
  ],
  "article_url": "https://www.washingtonpost.com/sports/football-rule-change"
}

---

📈 Tech Stack

Layer	Tools Used
Backend	FastAPI, Uvicorn, Pydantic
Frontend	Streamlit, Matplotlib, WordCloud
NLP Models	HuggingFace Transformers: Flan-T5, RoBERTa
Analysis	Pandas, Seaborn, ydata-profiling
LLM Hosting	Locally saved models in /models/

---

🔮 Future Enhancements

• 🔐 Add JWT authentication

• 🐳 Dockerize backend & frontend for deployment

• ☁️ Deploy on Render, Railway, or HuggingFace Spaces

• 📤 Export chat reports (PDF/CSV)

• 💬 Multi-transcript support with database

---

📄 License

MIT License © Yash Dewangan

⭐ Show Your Support

If you liked this project, give it a ⭐ star on GitHub and feel free to fork it!

💬 Let's Connect

Feel free to connect or suggest improvements!

• Built by Yash Dewangan
• 🐙Github: YashDewangan
• 📧Email: [email protected]
• 🔗Linkedin: YashDewangan

🤝 Contributing

Contributions, issues, and feature requests are welcome! Feel free to check the issues page (if you have one) or open a new issue to discuss changes. Pull requests are also appreciated.