[ AI integration ]
AI integration for websites and apps, only where it adds value
AI is worth it when it handles a specific task better or faster than the alternative. I integrate OpenAI, Google Gemini and Replicate models into websites, apps and systems, with costs and data handling clear from the start.
Where AI integrations usually go wrong
- 01
AI for the sake of it
A chatbot nobody asked for, or a flashy feature that solves nothing for the business. If it doesn't save time or improve the product, it doesn't belong.
- 02
The demo works, the product doesn't
An AI prototype is easy to put together. Shipping it means handling errors, slow model responses, usage limits and answers someone has to review.
- 03
Costs and data out of control
Every model call costs money and the data travels to an outside provider. Without a design that accounts for both, they get out of hand.
[ AI integration ]
AI integrations I build
Assistants that know your business
Chatbots and assistants that answer with your own information, such as FAQs, product catalog, policies or internal docs, and hand over to a person when they don't know.
Content and images
Copy, social posts and product images generated in your brand's style, with a review step before anything goes live.
Smart search
Search that understands what people mean, not just the exact words they typed.
Data from documents
Pulling information out of PDFs, invoices, forms or photos and loading it into your system, structured and without retyping.
Automation
Sorting inquiries, summarizing conversations or drafting replies so your team only has to review and approve.
Personalized proposals
Plans or recommendations built from each user's data, which a professional then reviews and adjusts.
AI inside real products
Landing + AI SaaS01Feedmakers
In Feedmakers, AI does the core work: you write one line and get the image with the text inside plus the caption, in the brand's style. Credits are only deducted when an image is generated successfully.
Landing + SaaS02Lucy
Lucy uses Replicate models to place each lamp in a real room, by day and lit at night. It's image generation applied to a specific need: showing how each product looks and how it lights a space.
Landing + SaaS03Parell
Parell generates weekly meal plans with calories and macros per meal, using Gemini and OpenAI behind its own interface, so switching providers doesn't touch the rest of the system. The nutritionist reviews and edits every plan.
Site + tools04ilove3d
ilove3d shows the other side: its tools process files in the browser and never upload them to a server, because with patient scans privacy comes first. AI 3D generation from a text description is listed as an upcoming tool.
How I work
- 01
Talk
We go over your business, your audience and what the site has to achieve. Then I send a proposal, timeline and budget.
- 02
Design
I design it in Figma with your brand and copy. We refine it together before any code.
- 03
Build
I build in incremental deliveries so you see real progress, on desktop and on mobile.
- 04
Launch
Publishing, domain, analytics and technical SEO. After that, maintenance if you need it.
[ FAQ ]
AI integration questions
There are two costs: development, and API usage, which the provider bills per request or image. Before building, we estimate expected usage and, if needed, add limits or credits so spending stays predictable.
Whatever is sent to a model goes through the provider's servers, so we send only what's needed and check their data terms before choosing one. If something can be solved without the data leaving your system, even better.
It can happen, so I design for it: answers grounded in your information, validation, and human review wherever a mistake matters. In Parell, for example, AI proposes the plan and the nutritionist decides.
It depends on the task, the cost and the quality you need. I prefer keeping the provider behind my own layer, as in Parell, so it can be swapped without rebuilding the product.
Yes. If your system has an API or an accessible database, it can be integrated without a rebuild. We look at that on the first call.
[ You might also like ]
Tell me which task you want to solve
Start with the problem, not the technology. I'll tell you whether AI is the best way to solve it and what it would take to do it well.