g a p p y

• 

  Our custom Raspberry Pi - powered LED ticker delivering real-time market insight

• 

  Example Dashboard Generated

Inspiration

In today's fast-paced financial markets, traders and investors struggle to make informed decisions based on vast amounts of data. Traditional trading tools often provide fragmented insights, requiring users to manually piece together historical performance, market trends, and sentiment analysis. We envisioned g a p p y as an all-in-one platform that leverages AI to automate and enhance the trading analysis process, making it more accessible and insightful for both novice and experienced investors.

What it does

g a p p y is an innovative, data-driven trading advisory platform designed to empower investors with personalized, actionable insights. The platform streamlines the entire trading analysis process, from data ingestion to trade recommendations. Users upload their trade history, which g a p p y processes to extract key details such as trade dates, ticker symbols, trade types, quantities, and monetary amounts. It then integrates real-time market data, performs clustering-based trade pattern recognition, and leverages sentiment analysis to generate tailored trading recommendations.

How we built it

g a p p y is powered by a robust backend built with Flask, handling data ingestion, analysis, and recommendation generation. The trade data is stored in a PostgreSQL database, while advanced analytics, including K-Means clustering, are used to identify trading patterns. The platform pulls real-time market data from sources like Yahoo Finance and processes financial news using embedding models stored in FAISS with HNSW indexing. Sentiment analysis is conducted using a fine-tuned FinBERT model, and AI-driven recommendations are generated using a retrieval-augmented generation (RAG) approach with a local generative model like Llama. The frontend presents insights through an intuitive, interactive dashboard with rich visualizations.

Challenges we ran into

One major challenge was ensuring efficient data processing and storage, especially when handling large trade datasets. Optimizing FAISS for quick vector retrieval and fine-tuning FinBERT for sentiment analysis required significant experimentation. Additionally, integrating multiple data sources seamlessly while maintaining low-latency responses posed technical hurdles.

Accomplishments that we're proud of

We successfully built a system that combines multiple financial data streams, AI-driven sentiment analysis, and personalized trade recommendations in a single platform. The implementation of FAISS for vectorized financial news storage and real-time sentiment evaluation was a significant achievement. Moreover, creating a user-friendly interface that distills complex data into actionable insights is something we’re particularly proud of.

What we learned

Through building g a p p y, we deepened our understanding of financial market data processing, sentiment analysis, and AI-driven trade recommendations. We gained experience optimizing machine learning pipelines for real-time inference and improving data efficiency using FAISS indexing. Additionally, we learned how to design a seamless user experience that makes complex trading insights easily digestible.

What's next for g a p p y

Moving forward, we plan to enhance g a p p y by incorporating reinforcement learning to simulate and predict optimal trading strategies. We aim to expand our market data integrations and introduce real-time alerts based on AI-driven insights. Additionally, we want to explore partnerships with brokerage firms to allow users to execute trades directly through the platform. Our long-term vision is to make g a p p y the go-to AI-powered assistant for intelligent, data-driven investing.

Built With

• axios
• bert
• chart.js
• faiss
• flask
• javascript
• llama3.3:70b
• node.js
• pandas
• passlib
• postgresql
• python
• rag
• react.js
• scikit-learn
• tailwind
• typescript
• yfinance