Resources

Field notes from real training runs

Practical writing on fine-tuning, data, evaluation, and serving, based on what we've measured rather than what's fashionable.

LLM Fine-TuningComing soon
Fine-tune, RAG, or both? A decision tree for enterprise teams
Five questions that settle the debate before anyone writes a line of training code.
Oct 2026
LLM Fine-TuningComing soon
How many examples does a LoRA adapter really need?
Learning curves from twelve tasks, from 250 to 50,000 examples.
Oct 2026
Training Data & RLHFComing soon
Writing annotation rubrics your reviewers will agree on
The inter-rater agreement playbook we use before labelling at volume.
Sep 2026
Evaluation & Red-TeamingComing soon
Can you trust an LLM grader? Calibrating LLM-as-judge
How we check automated scoring against humans before it gates a release.
Sep 2026
Deployment & LLMOpsComing soon
Serving an 8B model for under a cent per thousand tokens
Quantisation, batching, and GPU choices, benchmarked with vLLM.
Aug 2026
Custom Model DevelopmentComing soon
Continued pretraining for regional languages
What worked, and what didn't, when teaching an open model a low-resource language.
Aug 2026
RAG & Agent IntegrationComing soon
Teaching a tuned model to cite its sources
Combining fine-tuning with retrieval so answers stay tied to the documents they came from.
Jul 2026

Have a question we haven't written about?

Ask our engineers directly. The best questions become future articles.

We reply within one business day (IST).