Antoine: god level ETH Global Mentor. Learn from all past submissions to build the future
Antoine is a prototype for a trustworthy, verifiable AI agent designed to accelerate developer innovation and solve the "beginner loop" in hackathons.
Every hackathon, countless builders face the same problems: information overload from sponsor bounties and a lack of historical context, leading to redundant projects. For sponsors, this results in a low return on investment. Antoine tackles this by providing a strategic partner that has learned from the entire history of past hackathon projects.
Instead of just giving answers, Antoine provides deep, non-obvious insights to help builders stand on the shoulders of giants and create more sophisticated solutions.
Crucially, Antoine is designed as a blueprint for trustworthy AI. Our core mission is to solve the "AI black box" problem by building a system where the agent's knowledge, logic, and actions can be independently verified on-chain, creating a foundation for high-stakes applications where trust is paramount.
Antoine is a Python-based AI agent built on a multi-layered, decentralized stack, with a clear separation between our data pipeline, agent logic, and frontend assets.
Core Agent & Sponsor Tech:
The heart of the project is an ASI uAgent, which handles all reasoning and communication. The agent is successfully registered on the ASI Almanac testnet via a public ngrok tunnel, making it discoverable and fulfilling the core requirement for the ASI bounty.
Data Pipeline & Backend:
After discovering that standard scraping tools were insufficient, we reverse-engineered the ETHGlobal website's internal API to manually curate a high-quality dataset of 33 past projects, which is stored in prague_projects.json.
A "Librarian" script uses LangChain and the OpenAI API to generate vector embeddings for all project descriptions and technical stacks.
This structured data and the corresponding embeddings are stored in a Supabase Postgres database with the pgvector extension, creating a powerful, searchable "long-term memory" for our agent.
Scaffolded Architecture for Trust:
The Graph: We have fully scaffolded a subgraph (schema, manifest, and mappings) that is ready for deployment. This is designed to make Antoine's knowledge base publicly verifiable.
Oasis & Hedera: Our roadmap includes deploying Antoine to an Oasis TEE for confidential compute and using Hedera for an immutable audit trail, completing our "trust stack."
Frontend:
We created two static HTML/CSS/JS files: a demo.html with a polished, pre-scripted conversation for video presentations, and a landing.html page with a live Supabase email capture form to build a community waitlist.