AI assistants
Conversational interfaces trained on your product, policies and tone of voice.
- Engagement
- Fixed-scope build with a supervised review period
- Typical timeline
- 5 weeks
- Starts with
- An export of the last month of questions your team asked each other.
Your team already has the answers, spread across a product catalogue, a policy PDF, a pricing sheet and four years of email, so finding one takes longer than using it. An assistant only helps here if it says it does not know as readily as it answers.
Instant answers for staff and customers, in every language you sell in.
What we build
Concrete artefacts, handed over and documented.
- 01
A conversational interface inside the tools people already have open, your admin panel, your portal, Slack, rather than on a separate site nobody visits
- 02
Answers that quote the source sentence and link to the document, so anything important can be checked in one click
- 03
Explicit refusal behaviour: when nothing relevant is found, it says so and names the person to ask instead
- 04
A weekly review queue of the conversations it handled badly, which is where the following month’s improvement comes from
What changes
Answers arrive in the same window as the work instead of after a search
New staff reach useful independence in weeks rather than months
The questions nobody can answer become a visible list instead of a shrug
How it runs
01Question inventory
We collect what your team genuinely asks each other, from channels and inboxes, and sort it by frequency and by cost of getting it wrong.
02Grounding and tone
We connect the documents, tune how it speaks, and set the point at which it should stop guessing.
03Live with review
It goes live for one team while we read the transcripts daily and correct what it gets wrong.
Chosen per project. Named here so you can see the shape of it.
- Vercel AI SDK
- OpenAI
- pgvector
- Slack
Questions we get asked
Tell us the outcome, not the tooling.
Send us the situation you are in. We will tell you which discipline it belongs to, what we would do first and what it costs, including when the answer is to wait.