Inspiration
We saw how both beginners and experienced investors struggle to find the right financial information at the right time. They often have to switch between multiple apps and platforms for market data, research, documents, and productivity tasks. We built TaskMate to bring these capabilities together in one simple, conversational assistant.
What We Built
TaskMate is a finance assistant that understands a user's financial needs, works with their data, and can take actions on their behalf. It can track live stock prices, research companies, analyze documents and images, integrate with Google Drive, and understand voice messages — all through natural-language conversations.
How We Built It
We built TaskMate primarily with Python and used Gemini to speed up development and debugging. We divided the features between us, developed them independently, and then integrated everything into one assistant with a tool-based architecture.
We also added short-term and long-term memory, allowing TaskMate to maintain context and better understand user preferences and previous conversations.
Challenges & Learnings
One of our biggest challenges was prompt engineering. We iterated on our prompts many times to make the assistant behave reliably. Another challenge was handling messages containing multiple tasks, where the AI needed to correctly identify and execute multiple tool calls.
One lesson we learned was the importance of refactoring and designing for extensibility from the beginning.
What's Next
We designed TaskMate so new tools and capabilities can be added easily. Our next step is to refactor the code around SOLID principles and turn the core architecture into a reusable framework with common flows and tools. This could allow us to build hundreds of specialized AI agents by simply adding new prompts and tools on top of the same foundation.
Log in or sign up for Devpost to join the conversation.