Six years after my first hackathon in 2019, I jumped into my second — AdventureX 2025, held in Hangzhou, promoted as the largest ever in China.
Since arriving in China six months ago, I’ve checked out plenty of similar events, but most felt boxed-in, with rigid problem statements and little room for creative freedom. AdventureX stood out because it promised something rare in China, a truly open hackathon where you could build whatever you wanted. Honestly, I joined mainly in hopes of meeting sharp minds and seeing firsthand what others are building with AI right now.
Together with my team, I took inspiration from SUPCON’s vision of AI-run factories — where agents handle factory operations just by “talking” and reasoning like humans. We built “Agentic Factoria,” an AI system for real-world industrial needs, powered by SUPCON’s opensource supOS for its built in MQTT for real-time data handling. Agentic Factoria tackles factory management, fault fixes, and logistics, all using LLM-driven agents. Demoing this showed me just how powerful such systems can be.
Even after a year knee-deep in LLM applications, working on Agentic Factoria really challenged me to rethink context engineering. Keeping every agent on the same page with an immense amount of live data streaming in isn’t easy. We solved it with a shared order management system, short term states and long term history, system prompts for delineating roles, and output validation to keep everything tight.
There’s still plenty to improve, but getting noticed among 870+ people felt pretty awesome (we won first place in Moonshot AI’s Kimi track and second place in SUPCON’s track). On top of new skills, I made some great friends and learned where I still need to grow. I’m more motivated than ever to keep pushing the limits with LLMs and agentic tech — and to continue learning, both professionally and personally.