011 week
Use-case scope
We pick the one or two features with the clearest value, define what good looks like, and write the scope.
AI Integration
AI features inside a product you already have, in six to fourteen weeks, with evaluation and cost controls so the feature is trusted and affordable.
The buyer question this answers: We have a product. Can you add AI without breaking it?
Who this is for
- A product team that has prototyped with an API and needs it production-grade.
- A company with documents or data customers cannot find or use today.
- A workflow with a step a person does that a model could do with review.
Phase by phase
Reference timings for a typical engagement. The written scope gives you the dates for yours.
- 021 week
Evaluation set
A labelled test set of real inputs and expected outputs, built with your team.
- 033–8 weeks
Build
Retrieval, prompting, and integration into the product, with the score run on every change.
- 041–2 weeks
Guardrails and cost
Injection defences, sensitive-data handling, refusal behaviour, caching, and per-feature cost budgets.
- 051 week
Launch
Gradual rollout with monitoring of accuracy, latency, and spend.
- 0630 days
Support window
Fixes and tuning as real usage arrives.
The stack for this workflow
- Python
- Anthropic or OpenAI APIs
- Open-weight models where needed
- pgvector or Qdrant
- LangChain or LlamaIndex where they help
- PostgreSQL
- Docker
- Sentry
What could go wrong
Every build has risks. These are the ones we plan for on this kind of project, and what we do about each.
The feature is impressive in a demo and wrong in production.
How we handle itThe evaluation set comes before the build, and the score is reported weekly. Nothing launches below the threshold you set.
Running costs surprise you.
How we handle itCost per request is estimated in scope, measured in build, and budgeted per feature with alerts.
Users trust it too much.
How we handle itAnswers cite sources, the feature refuses when unsure, and the interface makes confidence visible.
Cost model: Fixed scope
One or two features with success criteria written down can be priced. Broader AI programmes suit a dedicated team.
No prices on this page. The scope phase ends with a written quote for your project.
From the blog
- LLM application architecture: the layers that don't change with the modelModel, harness, data layer, and product surface: the four layers of a production LLM application, and which ones survive the next model upgrade.
- RAG architecture: the pipeline from document to answerIngestion, chunking, embedding, retrieval, reranking, generation: the six stages of a production RAG pipeline, and where each one quietly determines answer quality.
- LangGraph vs CrewAI vs AutoGen: a practical comparisonThree popular agent frameworks, compared on what actually differs in production: control over the loop, debugging, and how much they decide for you.
Questions we get asked
Will it break our existing product?
It is built behind a feature flag with its own tests and rolled out gradually. Your existing paths do not change.
Which model will you use?
The one that scores best on your test set at an acceptable cost. The integration lets you swap models later.
What if accuracy is not good enough?
You will know by week three, from the score, and we will know why. Sometimes the answer is better retrieval, sometimes a narrower feature, occasionally not doing it. We say so.
Is our data safe?
API terms that exclude training on your data, or models running in your own cloud. Sensitive fields are filtered before anything reaches a model.
Tell us what you are building.
We reply within one business day with how we would build it, what it would cost, and which engagement model fits.
- 01Tell us what you are building
A short form or an email. No deck required, and "not sure yet" is a fine answer.
- 02A call with an engineer
Within one business day. Technical questions get technical answers, from the person who would build it.
- 03A written scope and quote
Fixed price where the scope is defined. The document is yours whether or not you go ahead.