Capstone
LocoLabo hosts student teams through Curtin’s Industry Capstone program. The lab brings a real problem, a working codebase, and the training scaffold. Students bring the analysis, the data work, and the evaluation.
Semester 2 2026 is the pilot: two briefs, one per track.
The two briefs
Section titled “The two briefs”Local AI versus cloud AI: a total cost of ownership model. A business team builds a model for when local inference beats a cloud API, and when it does not. The useful result is the crossover point, not a single headline number, and the model has to survive a sceptical finance committee.
Building and benchmarking a domain-specialist AI model. A technical team picks a professional domain, assembles a few hundred worked examples, and trains a small adapter on Qwen3-4B. Then they measure whether the specialist beats the base model on a sealed benchmark. A rigorous negative result is a valid outcome.
What students get
Section titled “What students get”- A scaffolded training notebook that runs on Colab’s free tier.
- Benchmarking time on lab hardware, arranged with the client.
- A defined contribution path into LocoLLM.
- A client for scope and data questions, and an academic supervisor for assessment.
Access to the briefs
Section titled “Access to the briefs”The capstone workspace, both briefs, the gate sheet, and the supervision pack are shared with allocated teams and supervisors. Ask the client for a copy.
Finished adapters follow the public LocoLLM contribution guide.
Contact
Section titled “Contact”Michael Borck is the client for both briefs. Scope, data, and benchmark questions go to him. Unit, report, and assessment questions go to the allocated academic supervisor.