Hammad Parveez - Software Engineer
Software Engineer. Mobile - Flutter, iOS, Android. Cloud - Firebase & Google Cloud. AI Agent Tooling & LLM Integrations. CI/CD & Release Automation.
I ship production mobile, web, and cloud systems, and build the AI agent workflows that help engineer them.
About Hammad Parveez
I build production software across the stack: Flutter apps on iOS and Android, TypeScript backends on Firebase and Google Cloud, and the CI/CD pipelines that ship them.
Lately I focus on AI-assisted engineering. At 10XU I built a GitHub workflow where an AI coding agent picks up issues on sandboxed runners and delivers reviewed pull requests, integrated LLM features into a consumer travel-wellness product, and published developer tooling like an MCP server for pub.dev.
Five years in, I have shipped recruitment, travel, and wellness apps to the App Store and Play Store, built real-time notification systems, designed a format-preserving encryption scheme for redemption codes, and written a few hundred test files along the way.
years of experience
commits at 10XU over 28 months
merged pull requests (97.5% merge rate)
AI-Assisted Engineering and Agent Workflows
I don't just use AI tools - I built the pipeline my team ships with.
GitHub Issue
@claude mentioned on the issue
running…
AI Agent
Claude Code on a sandboxed CI runner
queued
Pull Request
branch, commits, tests, checks
queued
Human Review
approved, merged, deployed
queued
›issue #214 opened — flight card shows stale status
›@claude mentioned by hammad
▌Rolled out across all company repositories: cross-repo pull requests, screenshot ingestion so the agent can read attached images, GitHub App auth, and a reusable dispatcher workflow. Dozens of agent-authored PRs merged after human review.
Shipped Work: iOS and Android App Screens
Real screens from apps on the App Store, personal projects, and developer tooling.
$ dart pub global activate architecture_cli
$ architecture_cli --create mvvm
✔ lib/models
✔ lib/views
✔ lib/view_models
✔ lib/services
guidelines written for every folder
8 patterns · MVC · MVVM · MVP · Clean · BLoC …
Work Experience
The timeline names the work. The systems behind the 10XU entry are broken down in Systems, with a diagram and the hard part for each.
10XU Limited (formerly WithU Holdings Limited)
Apr 2024 - Aug 2026Software Engineer · London, UK - Remote
- Core engineer on a consumer travel-wellness platform: Flutter iOS/Android app, a 33-domain TypeScript Cloud Functions backend, and four production web apps. 924 of 948 pull requests merged over 28 months, with commits across 24 repositories.
- Built and owned the company's AI coding-agent pipeline: a reusable cross-repo GitHub Actions dispatcher turning an issue mention into a reviewed pull request, hardened with GitHub App auth and defensive token handling. Adopted across every repository.
- Designed the redemption-code platform: uniqueness guaranteed by construction with FF1 format-preserving encryption, verified at one million codes, plus a redemption ledger and eleven admin endpoints.
- Built the flight and jet lag engine, the platform's largest work area at 119 pull requests: provider webhooks with automatic failover, IANA timezone and daylight-saving correct scheduling, and deduplicated push notifications.
- Built the video delivery pipeline: ffmpeg HLS and MP4 transcoding, cross-account S3 migration in parallel batches, CloudFront signed URLs, and an in-app HLS pre-cache layer for instant playback.
- Built the content pipeline moving workout and wellness content from Airtable through Firestore to S3 across three environments, wrapped in a single CLI so non-engineers could run a release.
- Owned iOS/Android release automation on self-hosted Mac runners: signing, store uploads, test gating, deploy notifications, and crash symbolication alerting into Slack. Ported the internal deal-admin tool from Flutter Web to Next.js in eight days.
Bayt (Talentera)
Jan 2023 - Apr 2024Senior Flutter Developer · Dubai, UAE - Remote
- End-to-end REST API integrations with offline caching (Hive, SharedPreferences), state management, and unit tests in a production recruitment app used by employers across the Gulf.
- CI pipelines (Codemagic, CircleCI) automating build, test, and release for Android and iOS, with custom flavours for several environments and teams.
- Cut latency in the app's critical user flows and refactored the codebase toward MVVM so further change was safe to make.
ShiftCTRL
Feb 2022 - Jan 2023Senior Flutter Developer · New York, USA - Remote, part-time
- Released the product on the App Store, Play Store, and Huawei AppGallery, and set up the build and release pipelines behind it.
- Built the app's REST integrations and offline storage (Hive, GetStorage, SharedPreferences), plus Firebase Cloud Messaging and local notifications.
- Extended third-party libraries in Kotlin and Java where the product needed behaviour they did not support, and led debugging and code review for the team.
Eusopht
Jan 2022 - Aug 2022Flutter Developer · Karachi, Pakistan
- Delivered apps through the full lifecycle from design to deployment; mentored junior developers.
Digital Graphiks
Jan 2021 - May 2021Flutter Developer (Intern) · Karachi, Pakistan
- Integrated WordPress REST, Shopify Storefront, and GraphQL APIs into Flutter apps.
Systems I Built and Owned
Production systems at 10XU, and the part of each one that was actually difficult.
Redemption Code Platform
One million codes, zero collisionsone-to-one mapping · collision retries: 0
Partner deal redemption end to end: code minting, a redemption ledger, eleven admin endpoints with paginated and batch reads, CSV export, and deep-link entry through Branch.io.
Uniqueness is guaranteed by construction rather than by retrying on collisions. FF1 format-preserving encryption is a bijection over the code space, so two deals cannot mint the same code. Verified against a one-million-code benchmark before rollout.
Video Delivery Pipeline
Segments pre-cached a step aheadsegments
signed
controllers pre-initialised one step ahead, memory watched so pre-caching never starves playback
Every workout and wellness video in the product moved through this: ffmpeg HLS and MP4 transcoding, cross-account S3 migration in parallel batches, and CloudFront signed URLs.
Story-style video had to start on tap, not buffer. An in-app HLS pre-cache layer fetches segments and pre-initialises controllers a step ahead of the viewer, with memory diagnostics so pre-caching never starves playback.
Flight and Jet Lag Engine
119 pull requests, the largest area14:20Z
tz
17:20 +03
deduped
handled
light, sleep and caffeine shifted per day toward the destination clock
The product's core loop: track a flight, generate a day-by-day light, sleep and caffeine plan around it, and notify the traveller at the right local moment. The largest single work area at 119 pull requests.
Almost every bug here was a timezone bug. Provider webhooks arrive in UTC, plans are lived in local time, and daylight-saving folds break naive arithmetic. It needed IANA-correct scheduling, deduplicated pushes, and a characterization test suite before the engine could safely be refactored.
Content Pipeline
Three environments, one commandgenerate
metadata
media
re-runnable, skips unchanged assets, cannot silently wipe published media
Workout and wellness content flowing from Airtable through Firestore to S3 across development, staging and production, with Dart generators and TypeScript uploaders behind a single CLI.
The people publishing content are not engineers. The whole thing had to collapse into one command that is safe to re-run, skips unchanged assets, and cannot silently wipe published media.
Venue Data Extraction
Accuracy measured, not assumedaccuracy scored against a hand-checked sample
City-scale gym and studio dataset built from third-party APIs: grid-based extraction over Google Places, automatic detection of each venue's booking provider, and LLM agents reading day-pass and pricing detail out of unstructured pages.
Scraped data is worthless if you cannot say how wrong it is. Accuracy was measured against a hand-checked ground-truth sample rather than assumed, which is what made the output usable for a product decision.
Product Analytics and KPI Reporting
One event map, three destinationsone definition
RudderStack in the middle, so a rename happens once instead of three times
A single event map wired through RudderStack into Mixpanel and BigQuery, plus automated daily and weekly engineering reports posted into Slack.
Getting product and engineering to trust one definition of each event. That meant a written event map, fixing an alias call that was merging Mixpanel accounts on login, and a volume flag to keep critical events readable.
Dart Service on Cloud Run
Same auth contract as the functionsJWT verify
same auth contract on both runtimes, so models are shared between app and backend
A Dart HTTP backend deployed to Cloud Run with Auth0 JWT verification, run as an alternative runtime alongside the TypeScript Cloud Functions.
Smaller than the others and honest about it: the interesting part was proving a Dart service could sit behind the same auth contract as the existing functions, so the team could share models across app and backend.
Open Source Projects and Developer Tooling
Published packages, AI tooling, and systems people can actually install.
$ python generate.py --module beginner_m1
voice: am_michael (Kokoro, CPU)
✔ step_01_1.m4a 24 kHz mono 64k 23 KB
✔ step_01_2.m4a 24 kHz mono 64k 26 KB
↻ step_02_1.m4a unchanged, skipped
offline, no per-word API cost
English With Aliyah
Language-learning app built solo end to end, with two offline ML pipelines: Kokoro text-to-speech on CPU generates every lesson's audio, and on-device speech recognition (Whisper tiny.en via sherpa-onnx) scores the learner speaking, with no network and no per-use API cost. Plus a data-driven Flutter app localised in three languages, a Next.js content CMS, 122 test files and CI on GitHub Actions.
{
"pub-dev": { "command": "npx",
"args": ["pub-dev-mcp-server@latest"] }
}
› list_pub_packages
› get_package_details
› get_github_readme
AI assistants can now read pub.dev
Pub.dev MCP Server
Model Context Protocol server that lets AI assistants search pub.dev packages and read their documentation. Published on npm with a hosted HTTP API, so any MCP-capable agent can use it with one config entry.
Crypto Analysis Agent
Real-time crypto analysis system: live price feeds for top coins, 10+ technical indicators on live charts, and multi-source news scraping with LLM sentiment analysis behind a FastAPI backend.
same prompt · three models
tokens/s · latency · cost — benchmarked per run
LLM Model Compare
Web harness that runs the same prompt against GPT, Claude, Gemini, and Llama side by side, with latency benchmarks and Playwright end-to-end tests. Keys stay session-only in the browser.
$ dart pub global activate architecture_cli
$ architecture_cli --create clean
✔ lib/data — repositories, sources
✔ lib/domain — entities, use cases
✔ lib/presentation — views, state
guidelines.md written in every folder
MVC · MVVM · MVP · Clean · BLoC · GetX …
architecture_cli
CLI on pub.dev that scaffolds Flutter projects in eight architecture patterns - MVC, MVVM, MVP, Clean, BLoC and more - and writes per-folder guideline docs so teams start consistent.
4242 4242 4242 4242
easy_card_scanner
Flutter plugin for on-device credit card scanning, published on pub.dev with fixes over the upstream scanner integration it builds on.
Things I Build for Myself
Extensions, agents and pipelines built to solve my own problems.
Google Meet Recorder
Chrome extension that captures, records and transcribes a Google Meet call from the browser, with no meeting bot joining the room.
Browser Context Capture
A Chrome extension paired with a VS Code extension that streams the text of your active tab into the workspace, so Copilot can use whatever documentation you are reading as context without copy-paste.
Chat to Pull Request Agent
Web chat front end over the Cursor Background Agents API: describe a change in conversation and the agent branches, commits and opens a pull request. GitHub OAuth with streamed progress over server-sent events.
Local TTS Content Pipeline
Offline speech generation for lesson audio: Kokoro running on CPU, a small Flask studio for auditioning voices, ffmpeg AAC encoding at 64 kbps mono, and a manifest so re-runs only regenerate text that changed.
On-Device Speech Recognition
Offline speech-to-text inside a Flutter app: a Whisper tiny.en model pulled down in a native background task that survives backgrounding and resumes rather than restarting, then run locally through sherpa-onnx. One shared in-flight download so a prefetch and a manual tap can never start two, with install checks, retry, and a repair path for a corrupt model.
Proof of concept, wired into the lesson speaking step
Agent Runtime Security Notes
Four reference guides on running coding agents safely: runtime setup, prompt-injection defence, multi-agent usage, and a security quick reference.
Written reference, not a shipped system
Explainer Render Pipeline
Turns a recorded take into a finished short: Whisper transcript to an edit timeline to a Remotion render, with loudness normalisation and speed-up applied at encode.
Advisory Firm Website
Marketing and lead-generation site for a boutique audit and advisory firm across three regions. Astro and Tailwind static build, Playwright coverage, tuned for SEO and speed.
Built with placeholder branding, pending client details
Technical Skills
Languages
Mobile
Cloud & Backend
AI & LLM
DevOps
Web & Testing
Contact
Open to interesting engineering problems. Let's talk.