Rahul Gupta · Senior iOS Engineer for Hire

Q: What kind of AI work do you actually do?

Two kinds, and they are different jobs. On-device machine learning: at Billion Hearts I built a pipeline that read a photo library, extracted face vectors and clustered them on the phone. Agent orchestration: I run seven repositories and four shipping products through a layer that derives each project's state from git and the App Store Connect API. If you need a model trained from scratch, I am the wrong person and will say so.

2014 → 2018 · Infosys, 7C Studio, Altair

Four years learning to ship other people's software.

I started on Income Tax Department systems at Infosys, moved to group chat over websockets at 7C Studio, then spent two years at Altair on WEYV, an app that played music and rendered magazines in the same session. I wrote the auto-scrolling lyrics view and the logic that deleted downloads when a subscription lapsed.

2018 → 2021 · MakeMyTrip, then Amazon

Senior Software Engineer II, then SDE II on Alexa inside Amazon Music.

At MakeMyTrip I rebuilt the My-biz signup flow and integrated the MYRA chatbot. At Amazon I rewrote how Alexa talked to the servers and how the client handled the directives coming back, added deep-linked audio playback between two apps through the EAAccessory APIs, and helped launch Alexa in new territories.

2021 → 2022 · Zomato

Engineering Manager, iOS.

I built dish-level recommendations off a customer's past ordering, shipped address sharing, and fixed the case where several saved addresses sit close enough together that the app has to ask which one you meant.

2022 → 2024 · PhonePe

Engineering lead. Three products, three teams, all of it inside a payments app.

I led Account Aggregator with four junior developers and shipped it in six months. It became the largest Account Aggregator in India by registered customers. Then Credit Score with two developers, then Consumer Lending with four, built as an independent micro-app on PhonePe's server-driven UI framework. Within months of launch we had disbursed over ₹100 crore.

Account Aggregator moves people's bank statements, so I wrote custom encryption for storage, secured the Core Data layer, and shipped the whole thing as a dynamic framework to keep the source out of reach of the wider organisation.

The habit I sell now started here. I wrote a Python script that read a Swift file and generated its unit-test template, and it doubled how fast every iOS developer in the organisation wrote tests. Automating my own job was the point long before there were agents to do it with.

2024 → 2025 · Billion Hearts

Core team on PicSee, where I shipped on-device machine learning and agentic AI in the same year.

I owned the receive, profile and connections modules, written in SwiftUI, UIKit, SwiftData and agentic AI. I also built the image-processing pipeline: a multithreaded system that read the photo library, found the faces, extracted the face vectors, and ran them through on-device clustering, Chinese Whispers among others, to group someone's photos by person. All of it ran on the phone.

This is the part of my history that matters most to anyone hiring for AI work. I built the plumbing before I orchestrated the models, so I know where these systems break.

2025 → now · Doxel

Senior Mobile Engineer.

360° cameras and intelligent systems, automating a construction industry that moves fifteen trillion dollars a year. I built the DataDog analytics in the app along with the dashboards that watch it for anomalies and stability drops. This is a contract, and it permits outside work.

Now

And in parallel, I run seven repositories and four shipping products by myself.

I built an operating layer above the repositories. One planning surface derives each project's real state from git and the App Store Connect API instead of from my memory, and it holds a short list of the things that would cost weeks to recover from.

It pays for itself on the things I forget. I had set a remote-config flag days earlier and moved on; the moment Apple approved the release, that flag switched a feature on for every user at first launch. The layer told me that morning the feature was already live, while I could still do something about it.

7 repositories 4 products on the App Store 1 engineer 5/5 across 52 advisory sessions

What I've built, and what it costs me to keep it running.

Four products, one probe service, and a video pipeline. I own each one end to end: the code, the store listing, the privacy policy, the support page, and the maintenance after launch.

PixelPlay. IPTV player for M3U, Xtream and EPG. The one with real users. On the App Store · 1.1.8

Takewell. Video teleprompter that cleans the audio it records. In review

Postural. Posture scoring from a single photograph, on device. In review

Perch. Home Wi-Fi doctor. A Mac agent and an iPhone companion. In development pixelplay-health. A probe service on GitHub Actions that tells PixelPlay which channels are dead. Public repo, public run history. Go and read it. Unattended

Nothing ships because a model said so.

I run a private pipeline that assembles short video from generated and sourced footage. Everything it makes stops at me before it goes anywhere. In the last batch it produced five videos and I killed three of them.

Those numbers measure the gate rather than the pipeline. Generated work reaching production because nobody checked is the failure I am paid to prevent, and the only defence is someone willing to throw away output they have already paid for.

Ask anyone selling you AI work what their reject rate is. A person without a number has no gate.

Three ways this works.

I am on a contract and take a bounded amount of outside work. Expect me to turn things down.

AI-orchestrated delivery

You have something to build and no appetite for a team. I bring the orchestration layer with me and ship on the same footing I ship my own products: the store submission, the review rejections, and the maintenance nobody quotes for.

Advisory, for teams that already have engineers

Your people are capable and the AI tooling is not landing. What to automate, what to leave alone, where the output needs a human gate, and how to keep it from producing confident nonsense at volume.

End-to-end iOS builds

Idea to App Store to maintained. Swift and SwiftUI, on-device ML where it earns its place, StoreKit, fastlane, and App Review. I have been through that last one enough times to have opinions.

Questions people ask before they book.

The ones that come up on almost every first call, answered here so the call can start further along.

Can you take on work while you are employed?

Yes. My current role at Doxel is a contract that permits outside work. I take a bounded amount of it, which is why the booking form asks about timeline and budget before we speak rather than after.

Do you build iOS apps end to end, including App Store submission?

Yes, and the submission is the part most people underestimate. I have taken four products through App Review, written the store listings, the privacy policies and the support pages, and handled the rejections when they came back. Swift and SwiftUI, StoreKit for purchases, fastlane for the release pipeline.

What kind of AI work do you actually do?

Two kinds, and they are different jobs. On-device machine learning: at Billion Hearts I built a pipeline that read a photo library, extracted face vectors and clustered them on the phone. Agent orchestration: I run seven repositories and four shipping products through a layer that derives each project's state from git and the App Store Connect API.

If you need a model trained from scratch, I am the wrong person and will say so.

Which time zones do you work in?

I am in Bengaluru, on IST. Calls run 13:00 to 18:00 IST, which is 09:30 to 14:30 in Berlin and 08:30 to 13:30 in London, so Europe and the UK work without either of us keeping odd hours.

North America needs a late-evening slot arranged separately. Australia does not currently work at all, because their business day sits inside my contracted hours.

Do you work with teams that already have iOS engineers?

Often, and that engagement is advisory rather than delivery. Which parts of the workflow are worth handing to agents, where the output needs a human gate before it reaches a repository, and how to stop quality sliding as volume goes up.

What is your stack?

Swift and SwiftUI first, UIKit where it still wins, The Composable Architecture, SwiftData and Core Data, StoreKit 2, fastlane and XCTest. Enough Python and Node to build the tooling around a project.

At PhonePe I shipped custom encryption for financial data and built on a server-driven UI framework, so the awkward parts of a regulated mobile product are familiar.

What does an engagement cost?

There is no rate card, because the shape of the work moves the number far more than the hours do. The booking form asks for a budget band so twenty minutes is enough to find out whether we are in the same range.

I ship production iOS products end to end, alone.