06 — Service
AI Integration
Chatbots, tutors and assistants powered by modern language models, designed into the product rather than bolted on the side.
- Catalogues people cannot decode
- Repetitive support load
- An AI feature nobody clicks
Lahore, Pakistan · remote worldwide
SpeakLab has an AI tutor students actually practise speaking with. Ali Electronics has an assistant that answers "will this inverter run my house in load-shedding" in plain language instead of showing a spec sheet. Both are grounded in the business’s real data, and both are built to admit when they do not know.
What this fixes
- Right now
A customer cannot tell from a spec sheet whether the thing does what they need, so they leave to think about it.
Once it's builtAn assistant that answers the question they actually asked, in plain language, out of your real catalogue.
- Right now
A generic chatbot invents a price, and you either honour it or lose the customer explaining why not.
Once it's builtGrounded in your data with hard limits, and an honest "I don’t know — here’s who does" everywhere else.
- Right now
AI bills that arrive as a surprise at the end of a busy month.
Once it's builtKeys kept server-side, rate limits and a spend cap set before it ever goes live.
What you get
- 01An assistant grounded in your own content, not general internet knowledge
- 02Honest failure — "I don’t know, here’s who does" instead of a confident invention
- 03Streaming responses, so it feels instant rather than frozen
- 04API keys kept server-side, never shipped in the browser bundle
- 05Rate limiting and cost caps, so a bad day cannot produce a shocking bill
How it works
01Pick the job
One task it should do well. Assistants that try to do everything do nothing convincingly.
02Ground it
Your catalogue, documents or policies become what it answers from — a source, not a vibe.
03Constrain it
What it must refuse, what it must escalate, and how it says so.
04Ship and watch
Real conversations get read. That is how you find what it gets wrong.
Questions
Which AI model do you use?
Whichever fits the job and the budget — Claude, Gemini and the OpenAI models each win in different places, and cost varies by an order of magnitude between them. The integration is built so the model can be swapped without rewriting the product.
Will it make things up about my business?
That is the risk, and it is handled by grounding: it answers from your data, and where your data is silent it says so and points to a human. An assistant that invents a price has cost you a customer and possibly a refund.
What does an integration cost to build?
An assistant built into a site or app you already have starts at PKR 50,000 (about $180) — picking the one job it should do well, grounding it in your content, the constraints on what it must refuse, and the streaming interface it lives in. It costs more when your data has to be cleaned or converted before anything can be grounded on it, and I will tell you that during scoping rather than after.
How much does it cost to run?
Text assistants are cheap — usually a fraction of a cent per conversation, so a few hundred conversations a month is a rounding error. Costs get real when you process documents or images at volume, and that gets estimated honestly before you commit.
Is my data used to train someone’s model?
Not on business API tiers, which is what these are built on. That is a different arrangement from a free consumer chatbot, and it matters if your data is sensitive.
Other services
Let's talk
Tell me what you need ai integration for and I'll tell you what it takes.