AlphaForge - Multi Agent Framework for Trading

Inspiration

The financial markets are complex, noisy, and ever-changing—no single analyst or algorithm can capture the full picture. We were inspired by the idea of assembling a virtual team of AI agents, each with their own expertise, to collaborate, debate, and synthesize insights just like a world-class research desk. Our goal: bring explainable, robust, and automated decision-making to trading, leveraging the latest in agent orchestration and cloud technology.

What it does

AlphaForge is a multi-agent AI framework that analyzes stocks from every angle. It runs technical, sentiment, news, and fundamental analysis in parallel, simulates debates between bullish and bearish researchers, synthesizes risk perspectives, and produces a transparent, actionable trading decision. Every step is explainable, with feedback loops to catch hallucinations or unsupported claims, ensuring reliability and trust.

How we built it

We built AlphaForge using the Google Agent Development Kit and Google Cloud, orchestrating multiple specialized agents in a modular pipeline. Each agent is responsible for a specific domain—technical, sentiment, news, fundamentals, risk, trading, and reflection. We used LangGraph for debate simulation, and Streamlit for an interactive dashboard. The workflow is fully automated, scalable, and cloud-ready.

Challenges we ran into

Designing effective communication and debate protocols between agents.
Ensuring the reliability and explainability of AI-generated insights.
Integrating real-time data sources and handling noisy or missing data.
Managing orchestration and feedback loops to prevent hallucinations.
Balancing performance, modularity, and scalability in a cloud environment.

Accomplishments that we're proud of

Successfully simulating multi-agent debates and synthesizing their outputs.
Building a robust feedback mechanism to catch and correct hallucinations.
Creating a modular, extensible framework that can be adapted to new asset classes or data sources.
Delivering a visually appealing, user-friendly dashboard for end-to-end transparency.
Demonstrating the power of agent orchestration for real-world financial decision-making.

What we learned

Multi-agent systems can dramatically improve the depth and reliability of AI-driven research.
Debate and synthesis between agents lead to more nuanced, explainable outcomes.
Feedback loops are essential for catching errors and maintaining trust in AI systems.
Cloud-native design and modularity are key for scaling complex agent workflows.