Inspiration
The project was inspired by my passion for investing. Investment is a complex process with several workflows entailing data extraction (required at a frequent pace). What if the tasks of collecting data and processing it through a set of rules was pre-defined by the user basis risk profiles was automated and accessible to the users without going for expensive institutional grade softwares. This inspired me to build an AI system where multiple specialized agents coordinate to provide institutional-quality analysis accessible to everyone - from retail to institutional investors.
What it does
The system transforms natural language investment queries into institutional grade analysis. The Google ADK orchestrates 4 specialized AI agents (Stock Screening, Style Analysis, Portfolio Risk, Multi-Strategy) that coordinate automatically. The process works as below - User asks "find low-risk tech dividend stocks" → ADK routes to multiple agents → agents collaborate → user gets Bloomberg-quality analysis in seconds. ADK turns 4 separate tools into 1 intelligent investment advisor.
How we built it
Built using Google Cloud ADK framework with Python 3.12
- Created 4 specialized agents, each with domain expertise and Yahoo Finance API integration
- ADK handles multi-agent orchestration and query routing
- Deployed on Google Cloud Run with Docker containers for auto-scaling
- Used Gemini 1.5 Flash for natural language processing Key innovation: intelligent query parser that determines which agents to coordinate based on user intent, enabling seamless multi-agent workflows.
Challenges we ran into
Google ADK Web UI wasn't displaying tool results in main interface despite functions working perfectly. Our detailed Yahoo Finance analysis was hidden in debugging events instead of user-facing chat. Root cause that we found was ADK's default LLM summarization was overriding our formatted outputs. This was rectified using tool context configuration, skip_summarization flags, and ensuring proper result flow from agents to UI. Browser caching also required fresh sessions for testing.
Accomplishments that we're proud of
In this project the key achievement I would say is implementing genuine multi-agent coordination where four AI agents collaborate to give institutional-quality investment analysis with live market data that adapts basis user sophistication and knowledge of investing/portfolio analysis. Happy to have created a platform that transforms natural language questions into institutional investor quality insights accessible to everyone.
What we learned
i. User interface vs backend : Accurate intelligence in the backend is of no use if users can't access it on UI ; ADK's dafualt LLM summarization was a major learning that I struggled with ii. Managing live market data and cleaning /streamlining it is a key challenge - missing financials/ratios, limitations of the open source APIs, which needs to be addressed robustly to build a production grade system
What's next for Intelligent Investment Agent
I plan to enhance this system by adding the following to the existing 4 agents- i. Financial planning agent for goal based investing ii. Asset allocation agent to build a multi-asset portfolio basis risk profile of investor ii. Mutual fund screening agent Additionally, planning to make a system capable of capturing data from US as well as Indian markets
Log in or sign up for Devpost to join the conversation.