✨ Inspiration
We were inspired by the overwhelming complexity and fragmentation in the decentralized finance space. While DeFi offers powerful opportunities, most users—especially newcomers—struggle to understand where to start, what strategies to follow, or which protocols to trust. We imagined a future where interacting with DeFi could be as intuitive as talking to a financial assistant—one that listens to your goals and instantly delivers actionable, data-driven strategies. DeFiPilot is our first step toward realizing that vision.
🛠️ What it does
DeFiPilot acts as a voice-powered assistant that generates personalized DeFi investment strategies based on user intent. Users simply speak their goals—something like “I want a medium-risk Ethereum strategy”—and the app takes care of the rest. It parses the spoken input, enriches it using an intent understanding module, and fetches live APY and risk data from real protocols, including those in the Algorand ecosystem. Then, using Gemini AI, it creates a complete investment plan, complete with a strategy description and a risk/reward summary. This strategy is presented to the user, read aloud using ElevenLabs text-to-speech, and optionally turned into a personalized video via the Tavus API. All strategies are saved to the user’s history so they can be revisited, bookmarked, or deleted later.
🧱 How we built it
We built the frontend using React and TypeScript, styled with Tailwind CSS to ensure responsiveness and a clean UI. Voice input was handled using the Web Speech API for maximum compatibility and low overhead. On the backend, we used FastAPI to build a modular and performant API that communicates with Supabase for persistent storage. For AI, we initially used GPT-4 but switched to Gemini due to quota limitations, adapting our code to work seamlessly with the Gemini Flash model. Strategy enrichment and generation were handled in tandem, and Supabase logged all strategies and enriched intents in real time. Audio narration was provided by ElevenLabs, while Tavus handled personalized video generation. The frontend is deployed via Netlify and connects directly to the FastAPI backend, eliminating the need for Flask, which we phased out during development.
🧗 Challenges we ran into
One of our biggest challenges was managing the AI integration. Initially, we used OpenAI's GPT-4 model, but we hit quota limitations midway through the hackathon. This forced us to quickly adapt and shift the enrichment and generation pipelines to Google's Gemini API. Switching providers under a tight timeline while maintaining response structure and quality was difficult. We also ran into problems aligning live DeFi data with AI output—ensuring strategies were both realistic and personalized required careful orchestration between data fetchers and language models. Integrating Tavus also required trial and error to match the expected payload and extract usable video URLs. Finally, replacing Flask with FastAPI and ensuring compatibility with the frontend involved debugging some persistent CORS and routing issues.
🏆 Accomplishments that we're proud of
We’re proud that every component of DeFiPilot is powered by real-time, live integrations. There is no mock data or hardcoded output. We managed to architect a system that smoothly blends multiple APIs—AI, voice, DeFi, audio, and video—into a single user flow. Our backend is cleanly modular and ready for production-grade extensibility. Switching AI providers without breaking the app mid-development was a huge win, and the final product feels polished, fluid, and genuinely useful.
📚 What we learned
Building DeFiPilot taught us how to work with live blockchain data, how to fine-tune AI prompts for structured output, and how to manage fallbacks when third-party APIs behave unexpectedly. We learned how to parse and act on voice input effectively, how to structure a Supabase schema to log meaningful analytics, and how to design around latency when using services like Tavus and ElevenLabs. It also taught us how to iterate fast under pressure, replacing major components (like OpenAI and Flask) without derailing the overall architecture.
🚀 What’s next for DeFiPilot
We plan to integrate real authentication using Supabase Auth so users can manage their own strategies securely. We'll also expand personalization by allowing users to specify portfolio size, token preferences, and yield goals. We want to add support for more chains beyond Ethereum and Algorand and introduce richer data visualizations for APY history and risk tracking. In the longer term, we imagine DeFiPilot evolving into a trusted daily companion for DeFi investors—one that suggests, adjusts, and explains strategies through natural voice-based interaction.
Log in or sign up for Devpost to join the conversation.