​A one-day HackSprint for fine-tuning, reinforcement learning and AI sandboxes

​On 29 August, Daytona, Give(a)Go and Baseline are bringing Dublin builders together for a full day of experiments at the edge of model training and agent infrastructure.

​The challenge is simple: take a model or agent, give it a measurable task, and make it improve.

​You might fine-tune a small model on a focused dataset, build an evaluation loop, train an agent with reinforcement learning, or use isolated sandboxes to run many experiments safely in parallel. The day is about working systems, visible results and lessons earned by building.

​What you will explore

• ​Fine-tuning: prepare datasets, run training jobs, compare checkpoints and understand where specialised models beat prompting alone.

• ​Reinforcement learning: design rewards, run rollouts, evaluate behaviour and improve an agent through feedback.

• ​Sandboxes: give every experiment an isolated, reproducible computer with the dependencies, files and tools it needs.

• ​Parallel experimentation: fork environments, test multiple approaches at once and keep the strongest results.

​Sandboxes make this practical. Each training job, rollout or agent variant can run inside its own disposable environment, without contaminating the others. Teams can reproduce failures, compare results and scale experiments without spending the day rebuilding infrastructure.

​Every builder will receive Daytona credits to build with during the HackSprint.

​Project directions

• ​Evolutionary fine-tuning lab: run fine-tuning jobs across datasets and hyperparameters, score them automatically, then promote the strongest checkpoints.

• ​RL agent arena: create a task environment where agents attempt the same goal, receive rewards and improve over repeated rollouts.

• ​Self-improving coding agent: benchmark agent variants against real repository issues and tests, then use the results to refine prompts, tools, memory or models.

• ​Sandboxed evaluation harness: spin up a clean environment for every model or agent evaluation so each run is reproducible and safely isolated.

• ​Synthetic data factory: use agents to generate, critique and filter training examples, then measure whether the resulting dataset improves a small model.

​You can arrive with a team or join one on the day. Bring a laptop, a model or agent idea, and a problem you can measure.

​What the day looks like

• ​11:00 AM: Doors open

• ​11:30 AM: Introduction to Daytona, sandboxes and the challenge

• ​12:00 PM: Hacking begins

• ​1:00 PM: Lunch

• ​1:30 PM: Back to building

• ​4:45 PM: Demos and judging

• ​5:30 PM: Awards and close

​About Daytona

​Daytona builds secure, elastic infrastructure for running AI-generated code. Its sandboxes launch quickly, fork into parallel branches and snapshot mid-execution, giving agents and training workflows isolated computers they can create and discard on demand.

​Venue partner

​Baseline is an independent community and fund for early-stage founders in Ireland. They back technical founders with €100K first checks and create a home for ambitious builders in Dublin.

​Join the community

​Give(a)Go is Ireland’s largest community of builders and founders, with 2,500+ attendees across 45+ hands-on events. From AI filmmaking and agentic hackathons to builder days, we bring ambitious people together to turn new technology into working projects.

​Follow us on LinkedIn, Instagram, our website and GitHub.

​We will take photos and short video clips during the day for recap posts and marketing. By registering, you consent to being included. If you prefer not to appear, tell one of the hosts when you arrive.