AI PRODUCT ENGINEERING · IOS AND ANDROID
Five model providers. One interface. No outage the user ever sees.
Anyone can wire a chatbot into a Flutter screen. I build the version that holds up in production: every model provider behind a fallback, entitlement verified on the server, a cost per request you can put in front of your board. Twenty-seven of my apps are live on both stores. Twenty-five clients came back.
Retention
Of my last sixty-nine completed engagements, fifty-four came from clients who had already hired me once. 25 clients have come back in total.
That matters more than a client count, and for a specific reason: nobody re-hires a developer who needed managing. One UK client has commissioned eight projects with me across roughly nine months, including two of the products below, built for the same team a year apart. Another shipped both halves of a two-sided marketplace with me. The longest single engagement on my books ran twenty-four months.
If you're weighing a six- or twelve-month build, that's the record I'd ask about.
6 of 17 documented builds. Each one has a full write-up: the problem, the stack, and what actually shipped.
Flutter
Flutter
Flutter
Flutter
Flutter
Flutter
Picking between GPT, Gemini and Claude takes an afternoon. What decides whether the feature works is everything around it: how the input gets broken up, what context reaches the model, whether the output is checked before a user sees it. The Actors PA uses five different AI services, each chosen for one job, because one model being good at everything is a marketing claim rather than an engineering one.
The Actors PA runs on five AI services and each one has a backup, because a Whisper outage at eleven at night should not end an actor's rehearsal. I have never shipped an AI feature sitting on exactly one provider, and I would argue against it if you asked for it.
An answer that arrives in fourteen seconds at forty cents is a failed answer. I size the model to the job, cache what repeats, and stream by default. You get a number per request before the build starts, not after the first month's bill.
Every AI feature I build has a defined degraded state. The rest of the product keeps running, the user is told what happened in plain words, and nothing they entered is lost. An app that goes blank on a model timeout is a broken app, not an AI app.
Subscription state, entitlements, usage counted against a paid tier. If a value decides who gets paid or what unlocks, the client app asks for it rather than asserting it. With AI every call costs real money, so a spoofed entitlement is not a free feature, it is somebody else's bill.
I'm in Lahore, and I've been shipping paid work since 2021. Since then: more than a hundred and seventy projects. Some built from nothing, many of them features added to apps somebody else started, a fair number of them fixes on code I inherited. Twenty-seven are in the stores today.
Flutter and Dart are where I live. Supabase, Node and Postgres behind them, with Firebase where auth and push suit it; Stripe, Klarna and in-app purchase where money changes hands; maps and real-time state, which is most of what a two-sided marketplace actually is. Lately, multi-provider AI inside apps that are about something else: the last two products I shipped run five and four external model services, each with somewhere to fall when it fails.
I'd rather spend the first week arguing about the data model than the last month working around it. That has cost me work. It is also why one of those clients is on their eighth project with me. The longer version is here.
I work alone. You talk to the person writing the code. That's the constraint, and it's also the product: I take a limited number of projects because each one gets my full attention.
Two groups, labelled honestly. The first is what the seventeen case studies were actually built with; every one of them lists its exact stack, and no two are identical. The second is the AI infrastructure I work in, which is newer than the case studies and kept separate for that reason.
iOS & Android
One codebase, both stores. Every project on this site is built on it.
Language · half a decade
Long enough to know which patterns survive a second year of the product.
Postgres · Realtime · RLS
For the projects whose data model wants SQL, and row-level security in the database.
Relational data
Bookings, ledgers, inventory. Anything that must not double-count.
Backend · REST · WS
Where the values that decide who gets paid are calculated and held.
Auth · Push · Realtime
For the parts that genuinely suit it: sign in, push, a realtime feed. The money and the state that has to be correct sit elsewhere.
Payments · Connect
Split payouts on marketplaces, where money moves between two users.
StoreKit · Play Billing
Subscriptions verified server-side, against a receipt from each store.
Subscription state
One entitlement truth across iOS, Android and a lapsed renewal.
Location · Routing
Most of what a two-sided marketplace is once you strip the screens away.
Model providers
Chosen per feature, each behind a fallback, so one outage degrades a session and lets it finish.
Voice · Vision
Transcription, distinct per-role voices and scene generation in The Actors PA.
Chunking · Embeddings · Reranking
For when the answer has to come from the client's own documents rather than the model's memory. Most “the bot is wrong” problems are retrieval problems.
pgvector · Pinecone
Semantic search over a product's own data, sized to the project rather than to whichever vendor has the loudest docs.
Function calling · Scoped tools · Human checkpoints
Tightly scoped, every step logged, a person in the loop at known points. Full autonomy is where agents fail in production.
Model serving · Async
Where inference, evaluation and the work that does not belong inside the app lives.
Answer quality · Regression
Answer quality measured before release and tracked after it, so “the bot is wrong” becomes a number somebody can act on.
67 reviews from 27 clients, quoted as written. 14 of those clients came back for another project. Every one of the six below is one of them.
This freelancer has a wealth of knowledge and professionalism and has completed the 1st milestone of this project ahead of schedule and to a superior standard. Looking forward to the next milestone and working more and more in the future!. 5 Stars *****
Very prompt and professional and excellent attention to detail, really grateful for putting me at ease through the process, gaining clarity and going above and beyond to deliver what was asked for within a reasonable budget, and looking forward to working on the next stages here
Zubair did an excellent job on the initial review and planning phase of this project. He provided a thorough and detailed assessment of the app, clearly identifying key issues, areas for improvement, and potential risks. His communication was clear and professional throughout the process, and he was very responsive to questions and feedback. I especially appreciated his structured approach in breaking down the next steps and being open to discussion around the best direction for the project. Overall, he demonstrated strong technical understanding, proactive communication, and a high level of cooperation. Looking forward to continuing to the next phase of development.
I had the pleasure of working with Zubair M on a mobile app project for my company, and I couldn’t be happier with the results. As a South African entrepreneur, I needed someone who could understand my vision and bring it to life, and Zubair M exceeded my expectations in every way. From the outset, Zubair demonstrated an exceptional level of professionalism, creativity, and technical expertise. His ability to translate my ideas into a functional and aesthetically pleasing app was truly impressive. The app not only meets all my requirements but also has a user-friendly interface that my customers love.
Zubair is very easy to work with. He first met with us to talk through our app idea, gave some feedback on how to best approach certain aspects of the app based on his experience developing apps, and then broke down how we would be working together by milestones. I am very happy with the progress we’ve…
Zubair has been working on my projects for quite some time now , I could not recommend him more highly as throughout all stages he has gone above and beyond to complete the tasks I have requested
Read all 67 reviews: 4.9 average.
Pick the one that matches where you are. Scope and timeline confirmed on the first call.
Idea to both stores.
Interface, backend, integrations, subscriptions, submission, and the review cycle after it.
The feature your product is actually about.
Chat, voice, vision or recommendation, added to a live codebase without destabilising it.
What my longest clients actually do.
Not a support contract. A standing engineering seat.
A working build every week
installable, on your device
Thirty days of support after launch
included on every tier
Let's discuss how we can bring your vision to life.
Real answers about AI engineering, delivery, and what working together looks like.
Have a specific question not covered here?