Real Demo Code
KEY FEATURE !!!
- GENERATE Productivity App, not beautiful junk
- One-click Deployment (based on Vercel & TiDB Cloud).
- AI Inspiration, Refining Ideas.
- Just input your idea to Convex
- One-click AI-generated Full Project.
- auto publish change to TiDB Cloud / Vercel
- One click to Generate Doc/ NextJS API & Web / TiDB Data Service / Chatgpt Plugin ....
Here's our answer:
People doesn't need to master programming, but to focus on creative thinking.
During this Hackathon, we validated two pioneering features of this concept.
To make sure AI create Productivity App, NOT **beautiful junk
What it does
one click deployment our seed app
input your idea.
Wait AI generate everything.
- chatgpt plugin
- tidb data service
TiDB Cloud and Vercel will handle all devops work!
How we built it
- NextJS front/ API
- Prisma ORM
- TiDB Serverless Database
- TiDB Data Service AP
- OpenAI Gpt4, Vercel AI SDK
- Vercel. deploy
Challenges we ran into
See us TiDB + Vercel Deploy button? It's magic. guess how we found and build it?
To generate controllable and expected projects without falling into the illusion of producing mere toys, we have implemented several approaches:
Leveraging cutting-edge research: We incorporate insights from the latest academic papers, including techniques like "Chain-of-thought" and "ReACT," to ensure our AI remains grounded in real-world applications.
AI Agent Pair Programming: We have introduced a novel method where two AI agents engage in a dialogue, providing different perspectives on the task. This iterative process helps generate stable and reliable code.
E2E Testing as a Reward: During training, we use End-to-End (E2E) testing as a rewarding mechanism. Multiple iterations are performed to generate projects that pass the user's project proposal-based tests. This significantly enhances the reliability of the resulting projects.
By combining these methods, we aim to produce AI-generated projects that are controllable, dependable, and align with user expectations, going beyond the limitations of mere toys.
Also We have optimized the TiDB Data Service.
By creating our custom OpenAPI to Data Service Config tool, we have successfully enabled AI to learn how to write TiDB Data Service configurations.
Also We found 10+ bug for TiDB vercel integration and TiDB Data service
Also like our 2022 tidb hackathon project mirror*. we are preparing a paper. hoping some conf can accept us again :)
Two weekends in a row of almost all-nighters, so tired, please automatically ignore the grammer errors and typo