Live Website: https://fin-tsp-qtqp.vercel.app/

💹 Fin-TSP: Neural Time-Series Terminal

Fin-TSP (Financial Time Series Prediction) is a high-performance Neural Terminal designed for the Indian Stock Market (NSE/BSE). This project moves away from traditional 1D price-action models, treating market dynamics as a Computer Vision problem. By converting raw price action into 2D Spectrograms via Short-Time Fourier Transform (STFT), a Convolutional Neural Network (CNN) identifies hidden spectral patterns in volatility to predict next-day price movements.

---

🛠️ Technical Stack

Backend (The Engine)

• Framework: FastAPI (Python 3.11)
• Deep Learning: TensorFlow-CPU (Memory-optimized for 512MB RAM constraints)
• Signal Processing: SciPy (STFT spectral mapping)
• Data Ingestion: yFinance (Real-time NSE/BSE data)
• Deployment: Render (Web Service)

Frontend (The Terminal)

• Framework: Next.js 15 (App Router / Turbopack)
• Styling: Tailwind CSS (Neural-Dark Theme)
• Visuals: Chart.js, Lucide React, & Framer Motion (Glow-effects/Animations)
• Deployment: Vercel

---

🧠 Core Architecture & Pipeline

The prediction pipeline follows a rigorous four-stage process:

1. Normalization: Raw OHLC data is scaled to ensure feature consistency across 5,300+ supported assets.
2. STFT Mapping: 1D price time-series is transformed into a 60x60 spectral image, capturing both time and frequency domains.
3. CNN Inference:
   • Conv2D Layers: Slide kernels over the spectrogram to detect "edges" in volatility and spectral density.
   • MaxPool: Downsamples features to ensure translation invariance.
   • Dense Head: A linear head performs the final regression to output the Predicted T+1 Price.
4. Backtesting: The system compares predictions against historical truth to calculate Directional Accuracy metrics.

---

Screenshots

💻 Local Setup

Ensure you have the necessary build tools:

1. Clone the Repository:

   git clone [https://github.com/rohan-shyam/Fin-TSP.git](https://github.com/rohan-shyam/Fin-TSP.git)
   cd Fin-TSP

2. Backend Setup:

   cd backend
   python -m venv .venv
   source .venv/bin/activate
   pip install -r requirements.txt
   uvicorn app.main:app --reload

3. Frontend Setup:

   cd ../frontend
   npm install
   npm run dev

---

📈 Roadmap

• STFT Spectrogram Integration
• CNN Model Training & Inference
• Real-time NSE Ticker Search
• Vercel & Render Production Deployment

Author: Rohan Shyam