Jalebi

Problem Statement

DeFi today feels like a jalebi — round, twisted, and messy. So we built Jalebi, an AI voice assistant for decentralized finance that makes interacting with DeFi as simple as saying what you want.At its core, Jalebi combines speech recognition, intent parsing, and on-chain automation to enable a new kind of DeFi experience — one where users don’t need to know which chain, which DEX, or which bridge to use. They just speak their intent, and Jalebi does the rest.Under the hood, Jalebi uses OpenAI’s Whisper model for speech-to-text (STT) and text-to-speech (TTS) conversion, allowing seamless voice interaction in the browser. The frontend is built with Next.js and TypeScript, offering a smooth and responsive interface that runs directly in the browser without any installation.For on-chain operations, Jalebi leverages LI.FI’s cross-chain architecture, enabling token swaps and bridges across multiple EVM-compatible chains. While we haven’t built our own DEX aggregator for the hackathon’s scope, LI.FI ensures deep liquidity access and reliable routing across major chains.Looking ahead, we plan to integrate the x402 protocol, transforming Jalebi into a fully autonomous DeFi execution agent. With this, users will be able to express their intent (e.g., “Bridge 100 USDC from Polygon to Arbitrum and stake it”), and the agent will autonomously handle:Token approvalsTransaction submissionsMulti-rail execution across same-chain swaps and cross-chain bridgesMonitoring and confirmation handlingDelivery of summarized execution outcomesAll write operations are protected with explicit user consent, slippage limits, minimum output guards, and scoped allowances to ensure safety and predictability.In short, Jalebi makes DeFi sweet, not sticky — bringing a voice-driven, intent-first, and user-friendly layer on top of complex DeFi operations.

Solution

Jalebi voice assistant is built with Next.js and TypeScript, following a modular, component-based architecture to keep the codebase clean, scalable, and easy to extend. Each component handles a specific part of the user journey — from capturing voice input to displaying swap and bridge results — allowing rapid iteration and easy debugging during the hackathon.We used React 18 for building dynamic UI components, and MUI for consistent, accessible, and responsive styling across the app. The application runs fully in the browser, with no backend dependency, ensuring quick interaction cycles and a seamless experience across devices.For cross-chain operations, we integrated Li.Fi’s SDK and widget to handle token swaps and bridging. This allowed us to plug into a proven multi-chain routing infrastructure while focusing our efforts on crafting a natural, intuitive AI-driven voice interface for DeFi.On the AI side, OpenAI’s API powers the natural language layer — translating user voice commands into actionable intents. Whisper is used for speech-to-text and text-to-speech conversions, making the voice interactions fluid and context-aware.Throughout development, we followed modern best practices — including strong type safety with TypeScript, efficient state management through React hooks, and clear separation of logic and presentation layers.The idea for Jalebi came from a personal frustration: DeFi feels unnecessarily complex for newcomers. I wanted to create something fun and futuristic — a system where you could simply say what you want to do and let an intelligent agent handle the rest.While the current version focuses on swaps and bridges, the future scope includes portfolio management, intent-based automation, and agent-driven transaction execution using protocols like x402.(And yes — AI helped us vibe-code a few late-night commits when caffeine ran low.) ☕🤖