AlphaPulse

Developer · February 2024

Problem

Financial signals are everywhere but scattered across dozens of sources. News sentiment, fundamentals, price history — you can look them all up, but pulling them together into an actual investment view takes time most people don't have.

What I Built

• 01Built an ingestion pipeline that scrapes and normalizes data from multiple financial sources: news articles, economic indicators, fundamentals, and price history. Everything lands in MongoDB so it's queryable in one place.
• 02The LLM layer does more than surface data. It synthesizes signals across all sources and generates structured reasoning: here's what the data shows, here's the context, here's what it might mean for the stock.
• 03The entire pipeline had to be built and deployed in 48 hours. Scraping, storage, LLM integration, and a live Streamlit interface, all running end-to-end by the time we demoed.

Financial data sources (news, fundamentals, prices)
  → Multi-source scraping + normalization
  → MongoDB (unified corpus store)
  → LLM synthesis layer (signal aggregation + rationale)
  → Real-time forecasts + structured recommendations
  → Streamlit interface

Results

First place out of hundreds of teams at Hacklytics 2024, Georgia Tech's annual 48-hour data science hackathon. The judges cited the breadth of data sourcing, the quality of the LLM reasoning, and the fact that we shipped a fully working product within the competition window.

Stack

Next Project