TradeSeer-AI

TradeSeer-AI is a multi-agent stock assistant using Google ADK to deliver real-time prices, news insights, and AI-powered forecasts with LSTM, GRU, Prophet—like a team of analysts at your fingertips.

Comment

🌟 Inspiration

As an investor and technologist, I constantly found myself switching between multiple platforms—Yahoo Finance for prices, Google News for headlines, and Python scripts for forecasts. I wanted a unified, conversational experience where I could simply ask, "What’s the forecast for AAPL using LSTM?" and get a smart, context-aware response. That’s where the idea for TradeSeer-AI was born: an intelligent multi-agent system that feels like a team of financial analysts available 24/7.

⚙️ What it does

TradeSeer-AI allows users to:

  • Look up real-time stock prices
  • Summarize the latest financial news
  • Forecast future stock prices using different models: Linear Regression, LSTM, GRU, and Prophet
  • Seamlessly choose forecasting methods through natural language prompts

All interactions are handled in a unified, agent-based conversational interface.

🏗️ How we built it

We used:

  • Google ADK to manage agents and tools
  • yfinance for fetching live stock data
  • scikit-learn, TensorFlow, and Prophet for implementing forecasting models
  • FastAPI for local deployment and testing

The root agent (named TradeSeer-AI) intelligently delegates tasks based on the user's query—using tools like stock_analyst, news_analyst, and prediction agents.

🧱 Challenges we ran into

  • Tool vs. Sub-Agent Decisions: Balancing modularity with performance—deciding what should be a Tool vs a SubAgent.
  • Forecast Accuracy vs. Speed: Training deep learning models (LSTM, GRU) on-the-fly while maintaining low latency.
  • User Intent Resolution: Building natural fallback prompts like “Which prediction method would you like to use?” when intent is ambiguous.

🏆 Accomplishments that we're proud of

  • Built a fully working multi-agent financial assistant using ADK
  • Integrated four unique forecasting methods into a single experience
  • Designed a natural fallback prompt to guide users when intent is unclear
  • Achieved consistent results across different tickers and forecasting horizons

📚 What we learned

  • How to structure a real-world multi-agent application using Google ADK
  • Time series forecasting best practices and trade-offs between models
  • Building intuitive interfaces for ambiguous or flexible natural language input
  • Managing cloud API quotas and latency in inference-heavy workflows

🔮 What's next for TradeSeer-AI

  • Deploy to Google Cloud Run for public accessibility
  • Add portfolio-level analysis and recommendations
  • Fine-tune models using longer historical datasets
  • Integrate with Google Sheets, BigQuery, or Vertex AI Forecast

Built With

Share this project:

Updates