Muneeb HussainAgent-first engineer

Agent-first engineer · taking on a few founders

My agents write the code. I own what ships.

I build AI products end to end: video dubbing, live captions, live voice translation, commerce and agent tools. My agents write most of the code. I design the system, review anything touching security or money, and ship it.

Built or led since July
16 projects
Commits since July
860+
Solo builds
3 AI media products
Say helloBook 30 minutes (opens in new window)

How I work

  1. 01You + me + agents

    Spec

    We agree what ships, what done means and what it costs. Agents draft the spec; we sharpen it.

  2. 02Agents

    Build

    Agents write most of the code, tests and docs against the spec.

  3. 03Me + agents

    Review

    Agents run tests and first-pass review. I read every change touching auth, data or money.

  4. 04Me + agents

    Ship

    Agents handle deploys and monitoring behind my approvals. Your team owns the code.

Work

860+ commits since July. Client work is described, not named.

BuiltMedia AI

AI video dubbing platform

Takes a video and dubs it into another language in a cloned voice, with optional lip sync and burned-in subtitles. Web, mobile and API.

How I built it

  • Temporal workflows run each stage: transcription, translation, voice, stem separation, timeline placement, lip sync. A routing step sends single- and multi-speaker videos down different pipelines.
  • Pipelines are locked and every output is stamped with its version. A job never switches provider halfway, so results stay consistent.
  • Lip sync is refused on multi-speaker or uncertain footage instead of shipping a bad result.
  • Top-ups are retry-safe and a worker recovers payments whose webhooks were missed. A per-minute cost model sits behind the pricing.

Taken from self-hosted GPU models to locked production pipelines, with about 50 test modules and a provider simulator for load tests that costs nothing to run.

  • FastAPI
  • Temporal
  • Postgres
  • Deepgram
  • ElevenLabs
  • Terraform
  • Expo
BuiltMedia AI

Live captions, transcription and translation SaaS

Batch transcription and translation for media teams, plus live translated captions burned into broadcast video feeds.

How I built it

  • Live captions hold a 30-second database lease, so a standby host takes over a stream within 30 seconds. Speech providers fail over in a chain.
  • Switched from CEA-608 to Unicode burn-in because 608 can't render Hebrew, Arabic or CJK.
  • Billing reserves minutes up front and releases them on cancel or failure. Translation has quality gates and retries weak segments on a second provider.

Seven transcription and seven translation providers behind one product, held to under 6 s of caption lag, with 38 test modules and failure drills in the runbook.

  • FastAPI
  • Celery
  • Redis
  • FFmpeg
  • S3
  • Stripe
  • Next.js
BuiltLive audio

Live sports commentary, translated by voice

Listens to live commentators, translates them and speaks the translation in a natural voice, out to viewers and studio software in real time.

How I built it

  • Translation waits for complete utterances, and each output track has its own queue, so multi-commentator rooms stay in order.
  • One continuous audio encoder per track replaced stitched MP3s. Previous lines are passed to the voice and translation calls to keep tone steady.
  • A circuit breaker guards translation providers, and the SRT output has a watchdog that restarts with capped backoff.

Delivered over HTTP, WebRTC and SRT. An end-to-end harness passed 13 of 13 workflows and caught 5 bugs before release.

  • TypeScript
  • Express
  • BullMQ
  • Deepgram
  • ElevenLabs
  • LiveKit
  • SRT
BuiltCommerce

Mobile-first fashion storefront

A womenswear store for shoppers arriving from Instagram ads on their phones, in Pakistan and the Gulf.

How I built it

  • The server re-prices every order and takes stock in one step, so two shoppers can't buy the last unit.
  • Cash-on-delivery orders are rate-limited per phone number and across the shop to stop fake orders.
  • An agent skill lets Claude run the store, with approval gates on production writes and an audit log.

Built solo end to end, with 67 backend tests.

  • Next.js
  • Convex
  • TypeScript
BuiltBroadcast

Broadcast scheduling integration

Moves media assets and metadata into a TV network's scheduling system through its business APIs.

How I built it

  • Every transfer is recorded before work starts, so re-sending an asset is always safe.
  • Runs as a dry run by default, so it could be tested against live systems without touching them.
  • Only updates records that already exist, and stops when a file name matches more than one record instead of guessing.

A safe bridge into a production broadcast system, built with Claude as co-author.

  • TypeScript
  • REST APIs

Also built and led since July

SecuritySecurity daemon for engineers' machinesBuilt

Runs on every engineer's machine and watches for compromise, supply-chain attacks and risky AI-agent setups, with a local dashboard.

  • 18 threat checks and 8 hardening checks, all local, with no telemetry.
  • CI runs on Linux, macOS and Windows with a coverage floor. No false positives across 76 repos on macOS.

The team can see exactly what each machine still needs to fix.

TypeScript · SQLite

InternalTamper-resistant team attendanceBuilt

Daily check-in for a team, with on-site or work-from-home decided automatically.

  • Passwordless login approved by an admin, with a short code sent over chat.
  • The server decides on-site status from the network; the browser can't fake its own IP.

Attendance the team can't game, from a written design spec.

Next.js · Convex

Consumer AIPalm and face reading appArchitected

A mobile web app that gives readings grounded in a deterministic Chinese-astrology engine.

  • The engine calculates and the AI only interprets, so readings can't drift.
  • Chart records are append-only; corrections create a new version.

288 of 288 fields matched across 24 reference charts, with tests running against a real database in CI.

Turborepo · Postgres · Drizzle

Lead genLead-generation pipelineRebuilt

Agents that scrape local businesses, plan outreach and write the messages.

  • Usage allowances are reserved inside a transaction to close an overspend race.
  • CSV exports are protected against spreadsheet formula injection.

Rebuilt from scratch into a deployable product with CI and single-server hosting.

TypeScript · Postgres · Stripe

SMSMissed-call text-back serviceHardened

Texts a customer a booking link the moment a local business misses their call.

  • Lead replies were being dropped. Added a masked-number two-way SMS relay with encrypted lead numbers.
  • Added magic-link sign-in, failed-payment handling and shared billing with a sister product.

Closed the gaps that were losing leads and payments.

Next.js · Twilio · Stripe

Voice agentsAI voice receptionist SaaSTech lead

Voice agents that answer a business's calls, in 18 languages, for multiple teams.

  • Owned phone-number provisioning, security hardening, billing and the free trial.
  • Set up the deploy and auto-merge pipeline, secret scanning and managed secrets.

The infrastructure, security and billing the rest of the team built on.

TypeScript · Twilio · Stripe

Agent toolsProposal generator for Claude CodeBuilt

Turns a JSON brief into a branded, paginated proposal PDF from inside an agent.

  • The build fails if content would overflow a page, instead of clipping it.
  • The agent must ask a person which price to quote; it never picks one itself.

Branded proposals generated from inside the agent, with 12 tests guarding the layout.

Python · Claude Code plugin

PersonalChat-first finance toolsBuilt

An expense tracker you talk to, and a personal ledger with a CLI and an agent skill.

  • LLM calls stay cheap by design, and period summaries are cached.
  • Every ledger change is logged with undo, and login rate limits can't lock the owner out.

Personal tools, built with the same safeguards as client work.

Expo · Convex · Next.js · DeepSeek

WellnessBreathwork practice platformTech lead

Guided breathwork and mindfulness practice organised around five principles.

  • Led and reviewed the build; a teammate wrote most of the code.

Shipped to a live deployment.

Next.js · Supabase

ExperienceSurah Yā Sīn, an interactive recitationBuilt

An interactive recitation experience of Surah Yā Sīn.

  • About 131k particles on WebGPU, with fallbacks for older devices.
  • The build fails if the Qur'anic text doesn't verify against its source.

A piece that can't ship with a wrong word.

Vite · WebGPU

This siteThe chat agent on this pageBuilt

Answers visitors' questions as me, captures leads and keeps every conversation.

  • Bans, rate limits and abuse filters run before the model is ever called.
  • Forged tool calls in user messages are dropped before storage, and admin login fails closed.

Try it with the button in the corner.

Next.js · Convex · AI SDK

Open source

Rates

Fixed prices. Tap a line to start a note with it.

Rate card2026

No hourly billing · equity only ever on top of cash

Every engagement includes

  • A one-page scope before any work starts
  • Direct access to me, no account managers
  • Progress you can see every week
  • Your team owns the code at handover

Not sure which line fits? Book a 30-minute call

Loadout

What’s in production with me this month.

Frontier models

  • Anthropic
  • OpenAI
  • Perplexity
  • Google Gemini
  • DeepSeek
  • Kimi
  • MiniMax

Agents & coding

  • Claude Code
  • Cursor
  • Codex
  • Opencode
  • Hermes Agent
  • OpenClaw
  • Pi Coding Agent
  • Manus

Cloud & infra

  • Vercel
  • AWS
  • GCP
  • Azure

Questions

The ones founders ask first.

Do you build it yourself?

Yes. Agents write most of the code. I design the system, direct the agents, review anything touching auth, data or money, test it and ship it. Your team owns the codebase after handover.

Is this vibe-coding?

No. Specs first, guardrails and evals, human review on anything touching auth, money or data, and deploys behind approvals.

Isn't AI just expensive hype right now?

Expensive for teams that use it carelessly. Used well, it's the cheapest senior engineer you'll ever hire. Knowing exactly where to spend tokens and where a cached call does the job is most of what I do. On my own team that works out to about $50 per engineer per month, and that's the whole bill. I optimise for what's coming, not what's trending, so you're not rebuilding it in six months.

Aren't you working full-time somewhere?

Yes, and that's the point. You get someone currently doing the job at scale, with evenings and weekends for async and a clear response-time SLA.

Do you take equity?

Sometimes, as a bonus on top of cash, never instead of it. If we both believe in the upside, great. But I don't trade my rate for a line on the cap table.

What if I actually need a full-time CTO?

I'll tell you on the first call. Disqualifying honestly is part of the service. A wrong engagement wastes your money and my time.

How fast can we start?

Discovery call this week, a one-page scope within a few days, and we can start the following week.

What are you shipping next?

Tell me in one line. I reply personally, usually within a couple of days.