feat/discover
sit in the room first
I take the stakeholder call. Requirements, constraints, and a straight answer on whether it's feasible.
+ scope agreed in half an hour− found out in week three
senior full-stack engineer · 5 yrs · lahore, pk · remote
abdulrauf
typescript · python · go
shipped for teams in uk · us · jp · nl · no
Five years building production SaaS platforms — multi-tenant, AI-native, and load-tested before launch. One clear span from requirements to production alert: no intermediate supports, no handoff at any boundary.
see the work →git checkout -b feature/how-i-work
on main — a request arrives
Usually vague, usually urgent, usually described as small.
feat/shape
pick what fits, not what's fashionable
The stack depends on the load, the team who'll maintain it, and how fast it has to move. I've shipped production in TypeScript, Python, Go and PHP.
+ chosen for the actual load− resume-driven architecture
feat/build
agile, and actually iterative
Increments short enough to change direction. I write the review comments too, not just receive them.
+ short feedback loops− big-bang merges
feat/harden
find the breaking point on purpose
Unit and E2E coverage, load tests against real throughput, auth and access boundaries checked before launch.
+ capacity is a known number− launch-day discovery
feat/ship
containerized, automated, repeatable
CI/CD pipelines, infrastructure defined in code, staged releases. A deploy should be boring enough to do on a Thursday.
+ boring thursday deploys− manual release rituals
feat/observe
the part most people skip
Dashboards, distributed traces, error tracking, custom metrics, and alerting that actually reaches a human.
+ failures announce themselves− silent breakage
git merge --no-ff → main
And I'm still on it afterwards. Shipping is the middle of the job, not the end of it.
the modern half
agents amplify judgment. they don't supply it.
The breadth came first — four language ecosystems before any of this existed. The harness that compresses my estimates is written up in full: conventions, agent roles, the MCP servers actually wired in, and the gates nothing merges without.
signal
0%
production bugs eliminated after containerization + full observability rollout
0%+
of usebetty.ai owned — first commit through production release
0
countries whose products I've shipped — uk, us, japan, netherlands, norway, pk
0ms
end-to-end voice latency, twice — once on fully custom orchestration in 2022
selected work
open — contract or full-time · remote
let's
talk
rauf.idev@gmail.comclick to copy
archive — 7 projects
work
Mostly SaaS, mostly multi-tenant, mostly things that break loudly when they break. Voice AI built before the platforms existed, autonomous delivery logistics, Japanese government data pipelines, UK edtech, Dutch equity plans, and a Norwegian task platform that quietly holds millions of users.
something not here?
ask
rauf.idev@gmail.comclick to copyprocess
how i work
Two halves. The sequence a piece of work moves through, and the tooling that compresses the mechanical parts of it. The first half hasn't changed in five years. The second changed completely in the last two.
ai-assisted development — guardrails on
agents write code.
i still sign it.
The breadth came first. I was shipping across four language ecosystems before any of this existed — MERN, then Laravel and Vue, then Go and Python. Agents amplify judgment. They don't supply it. What follows compresses the mechanical half of a build. It does nothing for the half that depends on knowing what should be built.
where a build's time actually goes
as originally scoped15 units
with the harness9 units
Illustrative — the real ratio depends on the codebase. The estimate drops because the mechanical part collapsed, not because anything got skipped. Review and hardening grow in absolute terms.
what's actually wired up
Skills teach an agent how to do a repeatable job. MCP gives it a live connection to the system where that job happens. The value isn't the tool — it's the specific hour of the week it deletes.
CLAUDE.mdconventions
Architecture boundaries and an explicit list of things never to do — the brief I'd hand a new engineer on day one.
removes: re-explaining how we do things, every sessionagent teamsseparation of duties
One drafts, one reviews against the spec, one runs the suite. No agent approves its own work — the same rule we apply to people, for the same reason.
removes: a single unreviewed opinion reaching the branchPlaywright + Chrome DevTools MCPbrowser
One drives the browser through the real user flow, the other diagnoses what it finds — network, console, performance.
removes: clicking through the same checkout for the twentieth timeContext7documentation
Version-specific library docs fetched at request time rather than recalled from training data.
removes: code written against an API that moved two releases agoSupabase / Postgres MCPschema
Migrations and types written against the tables that exist, not the ones somebody remembered.
removes: the migration that passes review and fails on a real columnSentry MCPproduction
Live errors with stack, release tag and frequency, pulled into the working context.
removes: guessing which deploy broke ithooks + eval gatesguardrails
Lint, types and tests block the merge. On retrieval work, a dropped eval score is a red build like any other.
removes: judgment calls that should have been a failing checkSkills — SKILL.mddeliverables
Packaged procedures, including docx / pptx / xlsx / pdf generation for specs and handover decks built from the same source of truth as the code.
removes: documentation drifting from the system it describeswhat none of this changes
- I read every diff before it merges, and I can explain any line of it in review with the tool closed.
- Architecture and data-model decisions are made before an agent is involved, by me.
- Tests are written against the requirement, not against whatever the implementation happened to do.
- If I can't defend it, it doesn't ship.
Fast is a result.
It isn't the method.
working together
three ways in
01 · solo
you need it built
No engineers to spare, or none who can take it end to end. I scope it, build it, deploy it, and stay on it after launch.
- discovery → release
- infrastructure included
- i own the deploy
02 · embedded
you need another senior
You have a team and a backlog that's outgrown it. I join properly — your process, your conventions, your review queue, not a parallel track.
- your standups
- your pr queue
- your conventions, not mine
03 · lead
you need someone steering
You have people but nobody scoping the work, estimating it honestly, or holding the review bar. That's a role, and I've done it.
- scoping & estimates
- code review
- mentoring
open — contract or full-time · remote
let's
talk
rauf.idev@gmail.comclick to copy
about
01 / 02 — click to swap
I started in JavaScript, on the MERN stack, building whatever the client needed. Five years later the list of languages is longer — but that was never the point. The point is that I stopped needing the stack to be decided for me.
Early on, "full stack" meant frontend, backend, database. That's a narrow definition and it stops being useful the moment something goes wrong in production. So the scope widened: CI/CD, infrastructure, scaling, monitoring, security. Not because I collected them, but because each one turned out to be the thing standing between a feature working on my machine and a feature working for real users at 2am.
What I actually do day to day is develop and ship. But I'm also the person who takes the call with stakeholders, turns a vague request into requirements, and says early whether it's feasible — instead of finding out in week three. I work in agile cycles: short iterations, real testing, PR reviews in both directions, and monitoring afterwards. End to end ownership, and I mean the end part.
I use agent teams, Claude Code and MCP in my daily loop, with guardrails, because velocity that skips review isn't velocity — it's debt arriving early.
the long version
how i actually build things
the whole stack, honestly
git log --graph --author=me
c0d1a04 2018 — 2022 · lahore
init — a degree and a job at the same time
B.S. Computer Science, University of South Asia, 3.9 CGPA. Started working professionally in the seventh semester, so the coursework and the production incidents overlapped.
7b3e441 2021 — 2022 · xeven solutions
feat: mern, and everything after it
MVPs through production apps, solo and in teams. REST APIs, OAuth, relational modeling — and the first time I had to think about a system rather than a screen.
2c88d10 2022 — 2023 · crewbloom · us
branch: gave software a voice
Owned more than half of usebetty.ai, a no-code AI call agent. Speech, telephony, queues, workers, and a canvas where non-engineers build their own call flows.
9d41ff7 2023 — 2024 · eukarya · jp
checkout: tokyo, open source first
Geospatial platforms for a Japanese government digitization programme. OCR pipelines at scale, medallion architecture, Go and Hono on GCP, all managed in Terraform.
b0e7a95 2024 — now · exam papers plus · uk
merge: senior, and the legacy problem
Several legacy systems merged into one Next.js and NestJS platform. Containerized, instrumented end to end, production bugs down roughly 99%. Then a RAG knowledge engine on top, with real evals.
f3c1e88 head → main
open to the next one
Strongest where AI systems meet infrastructure that isn't allowed to fall over.
open — contract or full-time · remote