project screenshot 1
project screenshot 2
project screenshot 3
project screenshot 4
project screenshot 5

FlexPI

FlexPI Redefine what API is by combining on-chain and off-chain data into a single, powerful solution. Create custom APIs with natural language, access real-time insights, and share or monetize your creations effortlessly. Simplify data, power innovation with FlexPI!

FlexPI

Created At

ETHGlobal Bangkok

Winner of

Blockscout - Blockscout Explorer Big Pool Prize

Prize Pool

Project Description

FlexPI (Flexible Programming Interface) introduces a fundamentally new approach to APIs. Moving beyond traditional request-response patterns, FlexPI combines AI with data orchestration to create APIs that understand context and adapt to needs. At its core, FlexPI connects blockchain (on-chain) and real-world (off-chain) data through an intelligent layer that processes natural language queries. This means developers can request data conversationally while FlexPI handles the complex integration work behind the scenes. The platform replaces rigid endpoints with "cognitive endpoints" - interfaces that learn from usage and automatically optimize themselves. Whether you're building crypto trading tools, managing portfolios, analyzing social sentiment, or monitoring DeFi protocols, FlexPI transforms how applications interact with data.

How it's Made

FlexPI's core architecture is built around our AI-powered Data Orchestrator, which leverages advanced Language Models and Reinforcement Learning techniques to seamlessly integrate on-chain and off-chain data sources. The system utilizes LangChain and LangGraph for sophisticated conversational AI workflows, combined with Retrieval Augmented Generation (RAG) to enhance context understanding and data accuracy. We've developed custom plugins for various data sources including The Graph, Blockscout API, Pyth Price Feeds, Dexscreener API, and Twitter, each implementing deterministic routing logic for query optimization. A notable technical achievement is our Graph Protocol integration, where we developed an efficient system to parse and utilize SDL (Schema Definition Language) files from subgraphs, enabling our AI to generate precise GraphQL queries from natural language inputs without requiring model training. The architecture is model-agnostic, supporting Llama, GPT, and Claude, with prompt engineering and few-shot learning techniques optimizing cross-chain data aggregation and transformation tasks.

background image mobile

Join the mailing list

Get the latest news and updates