🧠Narrative Nexus

A simple yet powerful web app built with FastAPI + Transformers + Vanilla JS that lets you:

• 🧹 Clean raw text or HTML content (using NLTK + BeautifulSoup)
• ✂️ Summarize the cleaned text using facebook/bart-large-cnn
• 💬 Analyze the sentiment (Positive / Negative / Neutral) of the generated summary

All in one neat, minimal dark-themed interface.

---

🚀 Features

• Drag & Drop Uploads — upload .txt or .html files directly
• Instant Cleaning — removes HTML tags, scripts, and unwanted formatting
• AI-Powered Summarization — compresses long text into key insights
• Sentiment Analysis — interprets the emotional tone of the text
• FastAPI Backend — lightweight and async
• Vanilla JS Frontend — no frameworks, just clean HTML + JS
• Offline-ready — supports loading models locally to avoid re-downloads

---

🧩 Project Structure

.
├── nexusnarrative/
│   ├── backend/
│   │   ├── main.py              # FastAPI entry point
│   │   ├── routes/
│   │   │   └── text_routes.py   # /clean-and-summarize endpoint
│   │   ├── text_process/
│   │   │   ├── cleaner.py       # Uses NLTK + BeautifulSoup
│   │   │   └── cleaned/         # Output directory
│   └── models/                  # (optional) local cached transformers models
│
├── frontend/
│   └── index.html               # Minimal UI
│
└── requirements.txt

---

⚙️ Backend Setup (FastAPI)

1️⃣ Create and activate a virtual environment

uv venv(intstall uv using pip install uv)

source .venv/bin/activate     # on Windows: .venv\Scripts\activate

2️⃣ Install dependencies

puv add uvicorn transformers torch beautifulsoup4 nltk

The first time you run the app, Hugging Face will download model weights (model.safetensors). These are cached locally and won’t re-download later.

3️⃣ Run the backend

uvicorn main:app --reload

The API should now be live at:

http://127.0.0.1:8000

Example endpoint:

POST /text/clean-and-summarize

It accepts a file upload (.txt or .html) and returns:

{
  "message": "File cleaned, summarized, and analyzed successfully!",
  "preview": "First 500 chars...",
  "summary": "AI-generated summary text...",
  "sentiment": {
    "label": "POSITIVE",
    "score": 0.987
  }
}

---

💻 Frontend Setup

1. Open frontend/index.html in your browser.

2. Make sure the API base URL in the file points to your FastAPI server:

   <code id="api-url">http://127.0.0.1:8000</code>

3. Upload or paste some text — and hit “Clean, Summarize & Analyze”.

---

🧠 How It Works

1. Uploaded text is cleaned with NLTK and BeautifulSoup

2. Cleaned text is summarized by BART (facebook/bart-large-cnn)

3. The summary is analyzed using DistilBERT sentiment model

4. The response includes:

   • Cleaned preview
   • Summarized text
   • Sentiment label + confidence score

The frontend displays all of these neatly in separate panels.

---

🗂 Example Output

Input: 3-page HTML article about global warming
→ Cleaned: 4,500 words
→ Summary: “Global emissions continue to rise as countries struggle to meet Paris targets...”
→ Sentiment: NEGATIVE (confidence: 98.7%)

---

🧰 Optional: Local Model Storage (Offline Mode)

If you don’t want the app to download models every time:

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, pipeline

model_name = "facebook/bart-large-cnn"
model = AutoModelForSeq2SeqLM.from_pretrained(model_name, cache_dir="./app/models")
tokenizer = AutoTokenizer.from_pretrained(model_name, cache_dir="./app/models")

summarizer = pipeline("summarization", model=model, tokenizer=tokenizer)

Then add: uv add "beautifulsoup4>=4.14.2" "fastapi>=0.120.0" "nltk>=3.9.2" "python-multipart>=0.0.20" "torch>=2.9.0" "transformers>=4.57.1" "uvicorn>=0.38.0" "huggingface-hub>=0.26.2"

export TRANSFORMERS_OFFLINE=1

---

🧑‍💻 Developer Notes

• Summarization input is truncated to 3000 chars to fit model limits
• Sentiment input limited to 512 tokens for performance
• Backend cleans up temporary files automatically
• Works smoothly with async FastAPI routes

---

🏁 Future Ideas

• Add translation support
• Multi-language sentiment detection
• Option to export cleaned + summarized text as .txt
• Docker support for easy deployment

---

🧡 Credits

Built with:

• FastAPI
• Hugging Face Transformers
• NLTK
• BeautifulSoup

Created by Akash Kumar (krakash2031@gmail.com)