We build LLM features into your product: an assistant, search over your documents, auto-fill, text extraction. We start with a pilot tied to one measurable KPI, then scale what works. Prices are public, and code and accounts are yours under the agreed terms.
Quote and schedule after your brief
We start with a free project brief and one scenario with a clear KPI, such as how many requests the assistant resolves without a human. Everything else waits until the pilot shows results.
We build RAG (retrieval: the model answers from your documents) and require source citations. If the data isn't there, the model says so and hands off to a human. Errors can't be removed entirely, but we measure them with tests.
We set per-user limits, cache repeated requests and match the model to the task: a cheap one for simple jobs, a stronger one for hard ones. Cost per request is visible on a dashboard.
We use provider APIs with training on your data turned off and mask personal data before sending. If needed, we host an open-weight model on your own infrastructure.
Use a package to describe what you need. We agree on the quote and schedule after reviewing your brief and requirements.
You want to test one AI feature on real users without a large budget
The pilot worked and you need a stable feature for all users
Several AI features, AI agents that take actions in your system, complex access rules
You describe the task and current workflow. We clarify requirements, choose a first stage and prepare a quote.
You provide a staging environment, sample documents and 20–50 examples of good answers. We create model provider accounts in your name.
We build a minimal version and test it on the query set and a small group of users. Payment is in stages.
We add guardrails, limits, logging and the final interface. We check the KPI and decide together what to scale.
You receive code, documentation and keys. Optionally we stay on monthly support to watch quality and cost.
We prepare a quote after reviewing your brief and requirements. A first stage can cover: project brief and scenario with a KPI; One feature: assistant, text summary or data extraction. Scope, payment and schedule are agreed before work starts.
We pick per task: OpenAI, Anthropic, Google or open-weight models. We structure the code so the model can be swapped without rewriting the feature.
Access to a staging environment, sample data or documents, and a person who signs off on results. Ideally 20–50 example questions with correct answers.
Requests go to the provider's API, but we turn off training on your data and mask personal data. If that isn't enough, we host an open-weight model on your own servers.
We build a test set of queries with expected answers and run it after every change. We also track the pilot KPI and review real user queries in the logs.
We agree on the handover of code, materials and account access before work starts. Third-party licences and services are listed separately when needed.
Describe your business, the outcome you need and your current workflow. We will clarify the requirements and prepare a quote after your brief.