01
New projects
A paragraph about the problem is enough to start. We reply within two working days, usually with questions rather than a quote.
AI · Service LLM fine-tuning and custom models
Fine-tune language models when prompting and RAG fall short: data preparation, training on open or hosted models, evaluation and deployment.
In short
Fine-tuning trains an existing language model further on your own examples, so it learns a format, style, classification or specialised task more reliably than instructions alone can achieve. It is not the usual way to give a model your knowledge — retrieval (RAG) does that better. We fine-tune when there is evidence it will beat a well-prompted model, and we prove it with evaluation.
- How to engage
- Product build
- Practice
- AI
Signs this is the right service
- A task needs a consistent output format or tone that prompting cannot hold.
- A smaller, cheaper or faster model must match a larger one on a narrow task.
- You need a model that runs in your own environment.
- You have thousands of good examples of the task being done well.
What you get
Is it worth it?
A comparison of prompting, RAG and fine-tuning on your task before any training.
Data preparation
Collecting, cleaning, de-duplicating and labelling examples, with personal data removed.
Training
Parameter-efficient fine-tuning such as LoRA on open models, or provider fine-tuning on hosted models.
Evaluation
The tuned model tested against the base model and the prompted baseline on held-out data.
Deployment
Serving in your cloud or through a provider, with monitoring and a rollback to the base model.
The process
- 01
Baseline first
The best prompted and retrieval-based result, so fine-tuning has something to beat.
- 02
Build the dataset
Quality matters more than size; we review samples with your experts.
- 03
Train and compare
Several runs compared on the same evaluation set.
- 04
Ship only if better
Deployed only when it measurably beats the baseline on cost, speed or quality.
LoRAHugging FaceOpen-weight modelsHosted fine-tuning APIsvLLMPython
LLM fine-tuning: common questions
RAG or fine-tuning?
- Use RAG to give a model knowledge — your documents and data, which change. Use fine-tuning to teach behaviour — a format, style or narrow skill. Many systems use both.
How much data do we need?
- For narrow tasks, hundreds to a few thousand high-quality examples are often enough with parameter-efficient methods. Quality and coverage matter more than volume.
Can we run the model ourselves?
- Yes. Open-weight models can be fine-tuned and served in your own cloud account, so data does not leave your environment.
Does fine-tuning reduce cost?
- It can: a fine-tuned small model may match a large model on one task at a fraction of the cost per request.
06 / Let’s build
Bring us the difficult problem.
02
Careers
Engineers, designers and researchers who like systems that are hard to get right. Send something you built and what you would change about it now.
03
Studio
Based in India and working remotely, with overlap hours agreed up front so your team always has a time to reach us.
04
Read
Guides on AI agents, RAG, MCP, tokenization and chain abstraction, written by the engineers who build them.