Last week, I had the pleasure of being part of the Code the Continuum Hackathon which took place at the University of Gagliari on Sardinia, Italy. The location itself is worth the visit, so I definitely understand the choice of venue. The two-day hackathon was held at the university campus with approximately 50 attendees registered. I had created a challenge where the participants would create a Jakarta EE application that integrated with an LLM to produce an Hourglass Model of a given domain and use case. The task was to give the LLM enough context to produce an accurate model using agentic augmentation techniques.
Five of the 10 teams signed up for the Jakarta EE Hourglass Model challenge out of which four of them finished. The teams were allowed to participate in multiple challenges, so there were some overlap between them. All teams did a fabolus job and it was not an easy task to select the three winners for our challenge. At the end, it boiled down to how well they were using the Jakarta EE APIs and the libraries provided to integrate with the LLM. And, of course, how accurate the generated Hourglass Model was.
The three winning teams were:
Cache & Release
AI/AB Users
Cyber Guys
The winning team, Team Cache & Release, created a generic solution where they incorporated guardrails for the quality of the generated model. They augmented the LLM with Tools using Jakarta EE and LangChain4J and generated a nice visual representation of the resulting Hourglass Model.
Team AI/AB Users got second place for their solution where they used Jakarta EE and MicoProfile in combination with LangChain4J-CDI to implement Tools, RAG and Guardrails for their model generation. They also created a visual representation of the model.
On third place, Team Cyberguys, used CDI Producers in combination with LangChain4J-CDI. They used advanced prompting to assure the quality of the model. They used metrics from a car to analyze and generate an Hourglass Model for the economy and safety of a driver. I discovered this functionality in the rental car I had, so it is definitely a relevant use case for SDV (Software Defined Vehicle).
