Custom-trained model
SPECIALIST UK CONSULTANCY · Professional services

A fine-tuned model that drafts client reports in the firm’s house style in 25 minutes instead of six hours.

First-draft time

~6h → ~25min

7 weeks · 1 PM + 2 engineers

Edits at first draft, not rewrites

~85%

House-style match (blind review)

~92%

Kick-off to production

7 weeks

Before

Where the team was when we picked this up.

  • Senior consultants were spending most of a day per report on first drafts.
  • Generic LLMs produced text the partners refused to send.
  • House style lived in three people’s heads.

What we built

Training set

Curated 1,200 historical reports with the senior team. Cleaned, structured, paired with brief metadata to make supervised training work.

Fine-tune + guardrails

Fine-tuned a smaller open model on the dataset. Added retrieval on top so client-specific facts come from records, not memory.

Editor in the loop

Senior consultants review, edit and approve. The model learns from accepted edits between cycles.

What changed

  • Consultants spend their time on analysis and client conversations again.
  • Reports went out faster and the partners stopped rewriting them.
  • New hires get a working draft to react to instead of a blank page.

After

Same team. Same week. Different shape of work.

Stack

Fine-tuned open modelAnthropic Claude (review)pgvectorPostgresCustom training pipeline

Timeline & team

7 weeks · 1 PM + 2 engineers

Got a workflow like this one?

Book a working session. We will tell you whether this is a four-week build or something bigger, and what it would take to ship it.

Book a working session

Cookies. Sadly not chocolate chip.

We use cookies to keep the site working, understand what is useful, and avoid shouting ads into the void. You can accept all, reject non-essential, or choose your own settings.

More detail lives in our Privacy Policy and Terms.