Inspiration
Our team didn’t start with a fixed idea. Instead, we explored different tracks, discussed multiple directions, and gradually narrowed things down based on our individual strengths and interests. After several rounds of pitching and brainstorming, we landed on Tiny Fish.
What drew us to this idea was not only its potential but also the opportunity to dive deeper into AI agents - something we were all curious about but hadn’t fully explored before. In the end, Tiny Fish felt like the most exciting and meaningful direction, both in terms of learning and building.
What it does
TLT is an AI-powered system designed to assist users in decision-making. Our system leverages nowcasting to identify and exploit the gap between real-time information updates and slower market adjustments on Polymarket. By detecting these temporary inefficiencies, it enables users to capitalize on short-lived mispricing opportunities.
How we built it
We built TLT by combining a frontend interface with an AI-driven backend system.
The frontend allows users to interact with the system easily and intuitively. The backend integrates AI models (such as LLMs or agent-based systems) to process input and generate outputs. We designed prompts and logic to guide the AI agent in producing structured and reliable responses.
Our workflow involved iterative testing: we continuously refined prompts, improved system logic, and adjusted the architecture based on results.
Challenges we ran into
One of the biggest challenges was working with AI agents and ensuring consistent outputs. Since AI models can be unpredictable, we had to spend a lot of time refining prompts and handling edge cases.
Another difficulty was aligning the system logic with real-world use cases - making sure the outputs were not only technically correct but also actually useful to users.
We also faced time constraints while learning new concepts, especially around AI agent design and integration.
Accomplishments that we're proud of
We are proud that we were able to build a working AI-driven system from scratch within a limited timeframe.
More importantly, we successfully explored and applied AI agent concepts, which was completely new to most of us.
We also managed to collaborate effectively as a team - leveraging each member’s strengths to overcome challenges and deliver a meaningful product.
What we learned
Through this project, we learned a lot about:
- How AI agents work and how to design prompts effectively
- The importance of iteration when working with AI systems
- How to turn a vague idea into a functional product
- Team collaboration under time pressure
This experience helped us better understand both the technical and practical sides of building AI-powered applications.
What's next for TLT
In the future, we want to improve TLT by:
- Enhancing the accuracy and reliability of the AI outputs
- Expanding its features to handle more complex use cases
- Improving the user interface for a smoother experience
- Potentially integrating real-time data sources
We believe TLT has the potential to grow into a more robust tool that can support users in real-world decision-making scenarios.
Log in or sign up for Devpost to join the conversation.