Inspiration
Web3 information is scattered across docs, X threads, and Discord servers. Evaluating whether a project is legitimate, understanding how a protocol actually works, or catching up on a fast-moving narrative (restaking, a chain's roadmap, a new L2) takes real research time — time most people don't have before they're already exposed to risk. We wanted a single place to ask Web3 questions in plain language and get a real, current answer.
What it does
BornAI is a Web3-focused AI research assistant with four modes: Research — deep dives into projects, protocols, and market narratives Explain — plain-language breakdowns of Web3 concepts (DeFi, NFTs, wallets, on-chain mechanics) Discover — surfacing relevant ecosystem trends and opportunities Analyze — structured due-diligence: risk signals, TVL trends, protocol health
How we built it
The frontend is React + Vite, deployed on Vercel. The backend is a single Vercel serverless function (/api/chat) that acts as a multi-provider AI gateway rather than depending on one AI vendor. It tries providers in order — DeepSeek, Anthropic Claude, OpenAI GPT, Google Gemini, and xAI Grok — and automatically falls through to the next one if a provider is rate-limited, out of credits, or erroring. Every attempt is logged with a structured event trail (provider_failed, provider_success, all_providers_failed) so failures are diagnosable instead of silent.
Challenges we ran into
This resilience design wasn't theoretical — it earned its place during development. Over the course of building this, we hit, in order: an API key environment-variable name mismatch, a JavaScript syntax error that silently broke the production build, an OpenAI account with no billing credit, a Gemini model version that got deprecated mid-build (forcing a model-name update), and a genuinely intermittent Gemini bug where the model's internal reasoning occasionally returned a malformed empty response. Each time, the fallback chain — plus structured logging that told us exactly which provider failed and why — let us diagnose and fix the real problem instead of guessing. It's also why DeepSeek ended up as the primary provider: it was the one with reliable, working credit by the time we needed a stable demo.
Accomplishments that we're proud of
Getting a genuinely resilient AI backend working end-to-end under real constraints, not just in a best-case demo — the app kept answering questions correctly even while individual AI providers were failing behind the scenes.
What we learned
Multi-provider fallback isn't just a nice architectural idea — real AI providers fail for mundane, unglamorous reasons (billing, deprecated model versions, intermittent bugs), and designing for that from the start makes the difference between a demo that breaks and one that doesn't.
What's next for BornAI
Expanding real-time on-chain data sources (contract verification, holder concentration, liquidity locks) so Analyze mode can back its answers with live data instead of the model's own reasoning alone, and adding persistent per-user history beyond the current browser-local storage.
Built With
- anthropic
- deepseek
- gemini
- github
- javascript
- node.js
- openai
- react
- vercel
- vite
Log in or sign up for Devpost to join the conversation.