---
title: StockSense · ML Dashboard
emoji: 📈
colorFrom: green
colorTo: blue
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: true
license: mit
short_description: LSTM + FinBERT stock prediction with Monte Carlo forecasting
---

# 📈 StockSense — Stock Price Prediction Dashboard

An end-to-end ML dashboard built by **Prabudh Rastogi**, Manipal University Jaipur.

## What it does

- **Live data** via `yfinance` — fetches real OHLCV prices
- **LSTM model** — Conv1D + Bidirectional LSTM with Huber loss
- **Technical indicators** — RSI, MACD, Bollinger Bands
- **7-day forecast** with Monte Carlo 80% confidence interval
- **FinBERT sentiment** — runs ProsusAI/finbert on live Yahoo Finance headlines

## Tech Stack

| Layer | Tools |
|-------|-------|
| Deep Learning | TensorFlow · Keras · Conv1D · BiLSTM |
| NLP / Sentiment | HuggingFace Transformers · FinBERT |
| Data | yfinance · Pandas · NumPy |
| Visualization | Plotly |
| UI / Deployment | Gradio · HuggingFace Spaces |
| ML Utilities | Scikit-learn · MinMaxScaler |

## Model Architecture

```
Input (seq_len=60, features=1)
  └─ Conv1D(64, kernel=3, relu, same padding)
       └─ MaxPooling1D(2)
            └─ BiLSTM(128, return_sequences=True) + Dropout(0.25)
                 └─ BiLSTM(64) + Dropout(0.25)
                      └─ Dense(32, relu)
                           └─ Dense(1)

Loss: Huber  |  Optimizer: Adam  |  LR Scheduler: ReduceLROnPlateau
Early Stopping: patience=5, restore_best_weights=True
```

## Resume Bullet

> **StockSense — Stock Price Prediction Dashboard**  |  *TensorFlow · HuggingFace · Plotly · Gradio*  
> Built and deployed an end-to-end stock forecasting app using Conv1D + Bi-LSTM; integrated FinBERT transformer for live news sentiment analysis and Monte Carlo simulation for 80% confidence intervals. Deployed on HuggingFace Spaces — [live demo]

---

> ⚠️ **Disclaimer:** Academic ML project. Not financial advice.