I turn napkin sketchesnapkin sketches into production apps.
What I do
Fourteen years shipping mobile products, and the backend and ML plumbing that keeps them alive after launch day.
At a glance
- shipping mobile products
- 14yrs
- live in the App Store & Play
- 10+apps
- notice for contract engagements
- ≤1week
When you can reach me
Mapping to your timezone…
Flexible either side for the right call.
- Your time
- --:--
- My time · Bengaluru
- --:--
- Of your 9-to-6
- – / 9h
Where I've shipped
- 2025 — nowNowOpen for a new build
Contracting for teams in the US, Canada and the Gulf.
- 2022Loop Health
Claims and live doctor consults in one React Native app.
- 2017Baker Hughes
Five years of industrial software at staff level.
- 2012Justdial
First release. 10+ store apps shipped since.
Teams I've built for10
Justdial, Xelpmoc, Shaw Academy, Baker Hughes, Loop Health, Expedia, Irish Life, Swiftsku, NCR Voyix, Alshaya Group
I don't just build apps. I built the AI-ready framework they ship on.
Companies that sell one app to many clients usually copy the codebase for each client, and every copy slowly drifts apart. Mobile App Framework ships one React Native codebase as many branded iOS and Android apps. It's AI-ready from day one: the architecture is enforced by tooling so AI coding agents can't erode it, and an in-app assistant can drive the same business logic as the UI.
- typed framework packages
- 13
- lines of framework TypeScript
- ~6k
- written architecture decisions
- 7
- lines of config in a new brand
- a few
Built for teams that ship with AI, in the codebase and in the app.
- For coding agentsGuardrails, not guidelines
Layer boundaries and import rules run in CI, so code written by Copilot, Cursor or Claude lands in the right place or fails the build.
- For LLM toolsTyped contracts everywhere
Strict TypeScript and zod schemas give models a precise surface to call, and reject bad output before it reaches a screen.
- For usersAn assistant with real hands
The in-app assistant calls the same actions as the buttons, asks before changing data, and is a per-brand feature flag.
Get things done, the Acme way.
- Buy milk
- Review the release notes
- Book flights
Coral palette, pill buttons, Arabic with right-to-left layout, AI assistant on.
Same code, different app
Switch the brand. Watch the app change.
Each app below runs the same screens and business logic. Only the brand package is different: name, bundle id, colours, shape, languages and feature flags.
export default defineBrand({
config: {
id: 'acme',
displayName: 'Acme Tasks',
i18n: { supportedLocales: ['en', 'ar'] },
features: { assistant: true },
},
theme: {
colors: { light: { primary: '#C2185B' } },
components: { button: { radius: 'pill' } },
},
});- Brand config
- a few lines
- Store bundle
- com.acme.tasks
What it covers
The parts a team ships with.
- White-labelA new brand is a config
Name, colours, languages and feature flags. The screens and the business logic stay put.
- ArchitectureThe structure is enforced
Layers are checked in CI, so the codebase stays clean as people and coding agents add to it.
- NetworkingRequests fail safely
Token refresh, retries and bad payloads become typed errors the screen can show.
- Design systemEvery brand stays on-brand
Colour, type and shape as tokens, light and dark, with contrast checked before release.
- LanguagesIncluding right to left
Copy and layout switch per brand. Arabic, French and English run the same screens.
- AI in the appThe assistant uses the real buttons
It calls the same business logic as the UI, and asks before it changes any data.
Agents that do real work, and ask before they touch anything.
I build LLM agents the way I build apps: typed tools, clear boundaries, tests in CI, and a human approval step wherever data changes.
- Tool useTyped tools, not prompt glue
Every action an agent can take is a schema-checked function with its own tests and permissions.
- MCPWired into the systems you run
Model Context Protocol servers for repos, issue trackers, analytics and internal APIs, so agents work from live context.
- OrchestrationPlanner, workers, reviewer
Multi-step and multi-agent flows with retries, budgets and checkpoints, so a long run resumes instead of restarting.
- Evals & guardrailsMeasured before it ships
Eval suites on real tasks, traced runs, cost and latency budgets, and approval gates on anything irreversible.
Agent project
Built · tested end to end · first live-model run nextBrand Agent
Type one sentence about a new client and Brand Agent sets up their version of my Mobile App Framework: name, colours, languages (Arabic, English, French, Spanish, Urdu) and features, with type-check, lint and every test green. Then it stops and waits for a person to approve before anything touches git.
The model decides, the harness enforces. Writes are locked to one folder (symlinks included), there is no commit tool, and approval is tied to a hash of the exact files that passed the checks.
- 7
- typed tools, zod-validated
- 1
- outside service: the model API
- 0
- ways to commit without a human
- 5
- shortcuts blocked in the e2e test
Kahwa, a Dubai coffee chain. Arabic and English, warm colours, loyalty assistant on.
Replay of the end-to-end test · model moves scripted, every tool ran for real
Everything you'd ask on the first call.
Screening details up front, and a builder that writes the intro email for you. No forms that vanish into a void.
- Status
- OpenContract or full-time
- Notice
- Within a weekFor contract engagements
- Based in
- Bengaluru, IndiaIST / UTC+5:30
- Setup
- Remote-firstOpen to onsite trips
- Experience
- 14 years9 companies, 10+ shipped apps
- Core
- React Native, Kotlin, JavaJavaScript, TypeScript, Node, Python, ML inference
- AI in the loop
- Claude, Copilot, CursorLLM-assisted build, review and prototyping — daily, not experimental
- Education
- M.Tech, BITS PilaniB.E. Computer Science, VTU
- Rate
- On requestDepends on scope and length
- Reply time
- Within the hourDuring working hours — email is fastest
Write the intro in 20 seconds
Pick what you're hiring for. I'll assemble a clean email — you just hit send.
Six things worth opening.
Mobile App Framework
My white-label React Native framework: one codebase shipped as many branded iOS and Android apps. The page shows how it works, the real errors its lint, types and tests return when you break the rules, a live brand switcher and the package architecture.
Explore the framework ↗Agentic AI · project pageBrand Agent
An LLM agent that onboards a new client onto my Mobile App Framework from a one-line brief, gets every check green, and waits for a human before committing. The page walks through the harness, the shortcuts it refuses, a live contrast lab and the eval briefs.
Open the project page ↗Product · liveA Dune
A remote job platform that pulls openings from every source into one searchable place. Built end to end — search, aggregation pipeline, and the front end people actually use.
adune.app ↗Healthcare · Play StoreLoop Health app
Insurance and care in one React Native app: claims, policy details, and real-time doctor consultations. Heavy focus on state correctness and keeping the app responsive on low-end Android.
View on Play Store ↗Logistics · Android4Tigo fleet platform
Native Android for truck fleet operators — live tracking, trip management and route work built around Maps, designed for drivers using it one-handed in a moving vehicle.
About the product ↗Tooling · open sourceAI visualization framework
Components for inspecting what a model is actually doing — inference traces and dataset shape rendered as something a team can read in a review, not a wall of logs.
See the repos ↗Fourteen years, eight teams, one habit: ship it properly.
- Full-stack delivery for product teams: mobile apps, APIs and the integrations between them.
- React Native, native Android and iOS work across insurance, finance and retail products.
- Performance and reliability work on apps already in the store — the unglamorous fixes that keep ratings up.
Depth first, breadth where it earns its place.
Mobile — where I'm deepest
- React Native
- Kotlin
- Java
- Swift
- Objective-C
- Native ⇄ RN migrations
- Push & OneSignal
- Huawei ecosystem
Web & front end
- React
- Next.js
- TypeScript
- Redux
- Tailwind
- GraphQL
Backend & data
- Node.js
- Python
- FastAPI
- Rust
- PostgreSQL
- MongoDB
- REST
AI & inference
- LLM integration
- Inference pipelines
- Latency tuning
- On-device models
- Deep learning basics
Build & release
- Fastlane
- Gradle
- GitHub Actions
- Jenkins
- Azure DevOps
- Firebase
- AWS
How I work
- Hands-on IC
- Ships own code daily
- Tech leadership
- Architecture reviews
- End-to-end ownership
- Agile / Scrum
- Unit testing
- Code review culture
The paperwork behind the practice.
Education
Certifications
Got a role that needs someone who's done this before?
Tell me the product and the timeline. I'll tell you straight whether I'm the right fit.
rizvan.g.h@gmail.com