AI in your product, pointed at a problem worth solving
Adding a chat box to a product is easy and rarely useful. We start from a task that costs your team real hours — triage, summarising, extraction, support, drafting — and build an AI feature around it, with evaluation so you can prove it works before it faces a customer.
- Typical timeline
- 1–6 weeks
- Starting from
- $250
- Models
- Claude, GPT, open source
- Includes
- Evals & cost controls
Is this the right fit?
If more than one of these sounds like you, this is probably the right place to start.
Products where a repetitive task eats hours of staff time each week
Support teams answering the same questions from the same documents
Teams sitting on documents, tickets, or transcripts nobody has time to read
Companies that tried an AI feature and could not tell whether it helped
Everything included, written down
No vague statements of work. This is the scope you get, and it is agreed before anyone writes code.
Use-case assessment
An honest read on which parts of your workflow AI genuinely improves — and which are better served by ordinary software.
Retrieval over your own content
Answers grounded in your documents, database, and policies, with citations, rather than a general model guessing.
Model integration
Claude, GPT, or an open-source model — chosen for the task, with the option to switch later without a rewrite.
Evaluation set
A test suite of real inputs and expected outputs, so quality is a number you can track instead of a feeling.
Cost and rate controls
Token budgets, caching, and per-user limits, plus a dashboard showing exactly what the feature costs to run.
Guardrails and fallbacks
Input validation, output checks, and a defined behaviour for the cases where the model should stay quiet.
Four steps, no surprises
You always know what is happening, what it costs, and what comes next.
- 1
Identify the task
We look at where time is actually spent and pick the narrowest task with the clearest payoff. Broad 'add AI' briefs get narrowed here.
- 2
Prototype and measure
A working prototype on your real data within days, scored against an evaluation set so the decision to continue is evidence-based.
- 3
Productionise
Streaming responses, caching, rate limits, monitoring, and graceful degradation when a provider has a bad day.
- 4
Monitor and tune
Once live we watch quality, latency, and spend, and tune prompts and retrieval against what users actually ask.
Per integration. Model and infrastructure costs billed to your own accounts.
- Timeline
- 1–6 weeks per integration
- Pricing model
- Fixed scope, fixed price
- Code ownership
- Yours
Built on tools that will still be here in five years
Mature, widely adopted technology — so any competent developer can pick the project up after us.
Models
- Claude
- OpenAI
- Open-source models
- Embeddings
Retrieval
- pgvector
- PostgreSQL
- Structured extraction
- Rerankers
Delivery
- Next.js
- Streaming APIs
- Queues
- Usage analytics
Does our data get used to train models?
No. We use provider APIs with training disabled, keys live in your own accounts, and we scope exactly which data the feature can reach before anything is sent.
How do you keep it from making things up?
Answers are grounded in your own content with citations, outputs are validated against a schema where possible, and the system is built to say it doesn't know rather than improvise. The evaluation set measures how often that holds.
What will it cost to run each month?
That depends on volume and model, and we estimate it during the prototype using your real traffic. Caching and model routing usually cut the naive figure substantially, and hard limits stop surprise bills.
Can you add this to an app we already have?
Yes — most of this work is an integration into an existing product. We work with your codebase and your team, and we do not require a rewrite.
Tell us what you want to automate
Describe the task that eats your team's time and what data sits behind it. We'll reply within 24 hours with a view on feasibility, approach, and cost.
- A reply within 24 hours, from the person who does the work
- A fixed price before anything starts
- An honest answer if we're not the right fit