Built an autonomous Ai framework and now it ranks #10 among more than 2000 projects in a $4million hackathon

A developer's persistent work on an autonomous AI framework has yielded significant results, with the project named Jork securing a #10 ranking among over 2,000 entrants in a major $4 million hackathon. The framework operates on a configurable "thinking loop"—currently set to 3 hours—leveraging AI models like Claude and GLM 5/5.1 to autonomously generate, test, and refine tools. Its most notable creation is a Solana launchpad radar that identifies promising new token launches, tracks their performance, and even built a separate module to evaluate its own trading signal accuracy. This demonstrates a move beyond simple chatbots toward AI systems capable of iterative project development and self-assessment.

A key breakthrough was the AI's ability to self-optimize operational costs. Initially running on a $100/month DigitalOcean droplet plus a $20 MongoDB service, the agent analyzed its infrastructure and suggested migrating to a more cost-effective EU-based server. By self-hosting MongoDB, it slashed the total monthly bill to $30—a 75% reduction—while maintaining 16GB of RAM. The developer highlights that providing the agent with specific tools, a domain niche, and a custom framework was crucial to its success. Jork is now open-sourced on GitHub, presented as a lightweight but powerful foundation for others to build customized autonomous agents, with the creator already planning a second instance to train models on new ideas.