Fine-tuning. When and how it ships.
Moonlabs is the operator-led AI Academy in Derby. We run three live companies — Homemove, home.co.uk and homedata.co.uk — and we teach twelve students per cohort to ship a real AI product, sell it to a real customer, and raise on it. Three pillars: Coding, Commercials, Investment. Twelve weeks. £6,000.
Moonlabs is what we are. Two operators — James Freestone and Louis O’Connell-Bristow — who run Homemove, home.co.uk and homedata.co.uk. We make the “fine-tune or just write a better prompt” decision in our own codebases every quarter — the discipline on this page is how we choose, not how we theorise.
The Academy is what we do. A twelve-week, in-person, twelve-student cohort in Derby. You build a real AI product. You sign a paid pilot on it. You write a deck and a financial model. You leave with a deployed system, a paying customer reference and a live investor pipeline. Coding, Commercials, Investment — the three pillars taught in equal weight every week.
Why this page exists. Most fine-tuning projects in 2026 should have been better prompts. Long context, structured outputs and tool use solve the majority of the use cases that needed a fine-tune two years ago. The first lesson is knowing when fine-tuning is genuinely the right answer; the second is doing it properly when it is — open-weight Llama, Mistral, Qwen with LoRA / QLoRA / PEFT, evals, deployment hygiene. You leave the Academy able to make the call honestly and ship the fine-tune competently — or as the founder of a vertical-AI product whose moat is its custom-tuned model.
Coding · production fine-tuning, end to end
LoRA, QLoRA, PEFT on Llama, Mistral, Qwen. Eval suites that compare a prompted baseline against the fine-tune honestly — the discipline that saves your team a month of GPU bills. Deployment patterns for self-hosted variants. Closed-model fine-tunes (OpenAI, Anthropic custom) where they earn it. A deployed fine-tune by week twelve, gated behind evals you wrote yourself.
Commercials · custom-models as a vertical-AI offer
A custom-tuned model trained on a customer’s proprietary data is one of the most defensible AI offers a small team can sell. Pricing the engagement (one-off + per-inference + maintenance retainer), the discovery call, a one-page pilot agreement. A paid pilot by week six — for many graduates, sold into a regulated industry that cannot use a frontier API directly.
Investment · raising on a custom-tuned-model moat
Cohere ($5.5bn on enterprise-tuned models), Mistral ($6bn), Aleph Alpha (sovereign AI fine-tunes), Reka, Together AI ($3.3bn on the fine-tuning infrastructure layer), Modal Labs, Replicate, Lambda Labs — the custom-models-as-moat thesis is one of the loudest funded categories in AI infrastructure. Cap table, ten-slide deck, financial model. A live investor pipeline by demo day.
Common questions.
Will I fine-tune Claude or GPT?
Closed-model fine-tuning (OpenAI fine-tunes, Anthropic custom models where available) is covered where the use case warrants it. The bulk of the practical fine-tuning work in the course is on open-weight models — that is where the leverage and the control sit for most production teams.
What hardware do I need?
For most LoRA / QLoRA work, a single modern GPU is enough. We provide access to the rigs we use ourselves for the heavier work during the cohort. You do not need to own an A100.
How does this fit with the wider LLM engineering curriculum?
Fine-tuning is one tool in the broader LLM engineering stack. The LLM engineering page is the wider framing; this page is the fine-tuning-specific entry point.
I have heard you should always start with prompting. Is fine-tuning dead?
No. Fine-tuning is alive for narrow, repeatable tasks where prompted approaches fall short and the latency / cost / privacy / control case is real. It is dead as a default first move, which is what most online tutorials still treat it as.
Will the course go out of date with new base models?
The tooling will. The disciplines — evals, baseline comparison, deployment hygiene — will not. We refresh the tooling each cohort.
More Academy entry points.
The Academy is one course with many doors. Each of these pages is a different entry point into the same twelve weeks.
Build it. Sell it. Raise on it. In twelve weeks.
Tell us where you are starting from and what you would fine-tune for — and whether the prompted baseline would actually do the job. James and Louis read every application personally and reply inside the week.
© 2026 Moonlabs Incubator. All rights reserved.