I wanted to share a project I’ve been working on a full-stack quant dashboard that predicts the next active market hour price direction of NVDA. Instead of relying strictly on price action, the goal was to build a pipeline that merges technical data with NLP sentiment. The backend (FastAPI) pulls live hourly candles (yfinance) and recent news headlines, running the text through FinBERT to generate a sentiment index. The backend chronologically merges the data, forward-fills missing sentiment gaps, and strictly calculates exact NASDAQ market states (handling weekends, pre-market, and after-hours shifts) to feed a clean, unbroken sequence of 10 data points into a PyTorch LSTM model. The frontend is built with Vue 3 and Chart.js, styled with a reactive cyberpunk theme because... why not. Obviously, this is purely an educational/engineering experiment and not actual financial advice. I’d love to hear your thoughts, especially any critiques on the PyTorch model architecture, the time-series merging logic, or how to better handle the real-time data pipeline. :)