Services
I build web apps with AI features and agents in them.
Chat interfaces, agent tools and background tasks that call a model, built into a real product with accounts, data and deployment.
Get in touch
Is this for you?
You run a product and want a chat interface, an assistant or an agent that works with your own data.
You have an idea for an AI product and need a first version that real users can try.
You've seen a demo that works and need it to work for many users, with accounts, limits and a bill you can predict.
Your team has the model part sorted and needs someone to build the web app and background jobs around it.
Training or fine-tuning models and heavy research sit outside what I do.
What you get
A narrow first version
We start by choosing one job the AI must do well, and how you will know it did. Most AI projects go wrong by starting too broad.
The AI part and the product part together
The interface, the call to the model, the tools it can use, the database and the sign-in. One person who holds the whole picture.
Background work that doesn't freeze the screen
Long tasks run in the background and report progress to the user, so a slow model call doesn't block the page.
Visibility and limits
Error reporting, usage tracking and application-side limits, with sensitive data kept out of logs.
Notes you can hand on
How the feature works, what it costs per use and what to watch.
What the first two weeks look like
Week 1
The job, the test and a skeleton
The job, the success test, a list of example inputs and the outputs you'd accept, and the product skeleton on a live address.
Week 2
A working first version
The first working version of the feature, behind a sign-in, tried against your examples, with logging turned on. After that we improve it using real examples from real use.
Relevant experience
On ScriptBee, an AI marketing platform, an agent run could take two to ten minutes. Inside a normal request, users who refreshed the page lost the answer. I moved chat work into Trigger.dev background tasks and built the Tasks page, so a run kept going and the user could come back to it. I also built two of the agent's tools, a Google Keyword Planner tool and a YouTube video analysis tool, plus the Google Ads connection.
At caisy, I added AI translation of content fields to a CMS editor. The ScriptBee case study has the details. I also work AI-assisted every day, using Claude Code and other AI coding tools.
How we work and what it costs
I build in TypeScript, React, Next.js and Node.js with PostgreSQL, and call models through their APIs. A scoped first version gets a fixed price, and ongoing work a monthly one. Model usage is billed by the provider to your account, so that cost is never hidden inside my invoice.
Use the form below to describe what the AI should do, who will use it and what data it works with.
Questions people ask
What kind of AI features do you build?
Features that call a hosted language model from inside a product, such as a chat interface, a tool the model can call, content translation or summaries, and long-running tasks that run in the background. I build them with the app around them, so they work with real accounts, data and billing.
Do you train or fine-tune models?
No. I build with hosted models through their APIs, which is enough for most products.
How do you keep AI features from costing too much or misbehaving?
Narrow the job, limit the input, log every call, set a spending cap and keep a person in the loop for anything that matters. What to build before you add AI to your product covers the checklist.
Do you use AI tools to write code?
Yes, every day, with Claude Code and other AI coding tools. The code still goes in through pull requests like any other change.
Related articles
What to build before you add AI to your product
A checklist for adding an LLM feature or an agent to a real product, from the narrow job to limits, logging and a person in the loop.
Other services
Take over a stuck app
The developer left, the prototype won't hold up with real users, or the team can't keep up. I read the code, tell you what is solid and what is risky, then ship the next release.
Build an MVP
One developer from first commit to launch. A small, working first version of your product that real users can try.
Build your web app or website
A site or app with a real back end, with accounts, a database, forms, payments and an admin side. Built, deployed and explained.
Arabic and bilingual web apps
Real right-to-left layouts, Arabic and English together, forms and numbers that behave. Built by a native Arabic speaker.