
Arfin NasirFull-Stack Engineer · Karachi · working worldwide
I build software that turns messy real-world input into decisions you can trust.
Stock tips in Roman Urdu, grocery orders across a city, trading signals checked against several sources. I take what's messy, check it properly, and make it run. 4+ years, 20+ projects, built in Karachi for clients worldwide.
New message in a stock-tips group
PASM khareedain 12.40-12.60 SL 11.80 TP 13.50 / 14.20
Read into a structured call
Verification slip
Checked in code
- Stock
- PASM
- Side
- Buy
- Entry
- 12.40–12.60
- Stop loss
- 11.80
- Targets
- 13.50 · 14.20
- Symbol is a real stock format
- Buy or sell is clear
- All prices are positive
- Stop sits below entry
- Targets sit above entry
- No hype words
Example message from my test suite, not real advice.
Verified call
- 4+ years shipping
- 20+ projects for 15+ clients
- 54 engineering articles
From noise to one signal you can trust
Everything comes in
Hundreds of messages, orders and alerts, in any language and any format. Most of it is noise.
The noise is filtered out
Cheap rules drop what can't matter before any AI is paid to read it.
The rest is read and lined up
AI turns each message into clear fields, and plain code checks every one of them.
One result you can act on
What survives is verified, with the reason attached. That's what reaches you.
Systems I've built, step by step
Each story follows one real input through a system I built. Open one to walk through it step by step, with an Engineer view for the technical detail.
- LiveRunning and verified in real use
- ShippedDelivered to a client
- BuiltCode and tests written, being verified in real use
- PlannedDesigned, not built yet
Finner · personal R&D
Reading stock tips from WhatsApp, and checking every one
Stock-tip groups on WhatsApp post calls all day in Roman Urdu. Some are noise, some are hype, and nobody keeps score of who is actually right.
The journey, step by step
- Built10 steps
- Planned2 steps
- 1Catch the message
- 2Skip the chatter
- 3Read it like a trader would
- 4Check it in code, not by AI
- 5Load real prices
- 6See what really happened
- 7Spot the red flags
- 8Score every group and sender
- 9Size it to my own money
- 10Keep an honest journal
- 11Alert within seconds
- 12Read screenshots and voice notes
- of messages never reach the AI
- ~65%
- AI call per message, and every number is checked in code
- 1
- unit tests on checks, outcomes, report cards, sizing and journal
- 55
- orders placed by the app: it informs, I decide
- 0
A fresh-grocery business in Pakistan
One system that runs a whole grocery business
Orders come from a website, a shop till, corporate clients and wholesale buyers. Every one has to be packed, driven and paid for without losing an item or a rupee.
The journey, step by step
- Built10 steps
Point at a step to see where it happens.

The delivery network
Main warehouse → van runs → dark stores → riders → homes
- Main warehouse
- Van run
- Dark store
- Rider
- Customer homes
Now on the map: A customer places an order
- API routes in one modular backend
- ~900
- database tables in one PostgreSQL database
- 236
- apps: five staff dashboards, a storefront and a rider app
- 7
- automated backend tests
- ~2,000
EFTS · client project
Trading signals, cross-checked before anyone acts
Traders get signals from many places at once. Acting on a single alert that nobody has checked is how money gets lost.
The journey, step by step
- Shipped6 steps
- 1A signal is captured
- 2Every signal is kept, exactly once
- 3Checked against other sources
- 4AI research, summarised
- 5The plan is checked
- 6The trader decides
- sources checked for each signal
- Several
- plain summary per signal, with sources
- 1
- of development
- 10 months
- engineers on the team
- 3
Khidmat AI · hackathon build
Booking a home-service pro from one sentence
Finding an AC technician or a plumber in Pakistan still means phone calls and guesswork. When someone cancels, you start all over again.
The journey, step by step
- Built8 steps
- 1Understand the request
- 2Find pros nearby
- 3Rank them fairly
- 4Book with a clear price
- 5The pro decides
- 6If they decline, re-book automatically
- 7Track the job
- 8Finish and rate
- AI agents working in sequence
- 7
- home-service categories
- 13
- languages: Urdu, Roman Urdu, English
- 3
- from idea to working prototype, team of two
- 4 days
Try it yourself
Paste a stock call, like the ones posted in WhatsApp groups, and watch the same checks run.
0/500
Try an example
A simplified, rules-only version running in your browser. Nothing you type is sent anywhere. The real reader uses AI first, then runs these same checks.
Verification slip
Checked in code
Paste a stock call to check it
Watch AI agents do the work
This is how I build agent systems, for bookings, trading and operations alike: each AI agent owns one job, uses real tools and data, hands the work on, and recovers by itself when something fails. A person still has the final say.
Example request, in Roman Urdu
AC kharab hai, urgent technician chahiye
A step fails: re-plan
Agent console
15/15
- Understandthinking: Roman Urdu: “AC kharab” means the AC is broken; “urgent” sets the priority.
- Understandresult: Job understoodservice: AC repair · urgency: urgent · confidence 0.9 (needs 0.7, so no question back)
- Searchtool call: Searches the map around the customergeo search · radius 10 km
- Searchresult: 6 AC pros found within 10 km
- Decidetool call: Scores every pro on six factorsdistance 25% · rating 20% · on time 15% · cancellations 15% · skill 15% · recent work 10%
- Decideresult: Top pick: Pro A, 2.1 km away, rated 4.8closest pro with a strong on-time record
- Actresult: Slot held for Pro A, price set for an urgent jobprice = base × complexity × urgency
- Confirmtool call: Asks Pro A's agent to check the calendar
- Confirmfailed check: Declined: fully booked this afternoon
- Recoverthinking: Drops Pro A, adds a small loyalty discount, plans again (1 of 5 tries)
- Recoverhand-off: Hands the job back to Decide, without Pro A
- Decideresult: Next best: Pro B, 3.4 km away, rated 4.7
- Actresult: Slot held for Pro B
- Confirmresult: Accepted by Pro B
- Reportdone: Customer and pro notified; the job is tracked until it's done
Booked from one sentence. Re-planned once. No phone calls.
An example run with demo data. The pattern, the checks and the recovery loop are the ones I use in production work.
See the systems I've builtWho you'll work with


I'm Arfin, a full-stack engineer based in Karachi. For 4+ years I've built web, mobile and backend systems for clients in Pakistan and abroad, from trading-signal research to the software behind a city-wide grocery operation.
I like the messy parts: WhatsApp messages, cash on delivery, people working in the field. You talk to me directly, from the first call to the handover.
Where I've worked
Jun 2024 – Present
Design District Digital
Full Stack Developer
Led dev teams, shipped App Store and Play Store releases, ran production infrastructure.
May 2023 – 2024
Ayan Tech
Full Stack Developer
REST APIs with MongoDB, React frontends on NestJS services, client delivery end to end.
Dec 2022 – Mar 2023
3 Const.
React Native Developer
Built and maintained React Native apps; debugged and optimized performance.
Mar 2022 – Nov 2022
HNH Softtech Solutions
React Native Intern
Prototyped apps from client requirements, tested components, joined requirement calls.
Arfin delivered exceptional work on our mobile applications. His expertise in iOS/Android development and backend development with Express.js and SQL DB was invaluable. We're very satisfied with the results.
What I'd build for you
The same patterns fit most businesses that run on messages, stock and people in the field.
Who I build for
- Shops and grocery delivery
- Traders and investment teams
- Home and field services
- Businesses that run on WhatsApp
- Startups shipping a first product
Messages into tasks
Orders, complaints or requests arriving on WhatsApp or email are read into structured jobs and checked before anyone acts on them.
One system for the whole operation
Stock, orders, staff and cash in one place, with every change traceable. No more twelve spreadsheets.
AI agents that finish the job
Assistants that understand a request, use your own tools and data, hand work between steps and recover when something fails, with a person approving what matters.
Field teams with proof
Riders, technicians or agents get an app that works offline, records proof of every job, and settles cash at the end of the day.
Ways to work with me
Build a product
A web or mobile app from first sketch to App Store and Play Store, including the parts that are easy to get wrong: payments, sign-in and real-time updates.
React, Next.js, React Native, Node.js, NestJS, FastAPI
Run your operations on one system
Orders, stock, staff and cash in one place, built around how your business already works. A setup fee, then a monthly care plan.
AI agents on your own data
Agents and assistants that read your documents and messages and act through your tools, with every answer checked or cited, not guessed.
Hiring full-time? I'm open to senior full-stack roles, remote or hybrid.
Download CV (PDF)How a project runs
- 1
A 20-minute call
You tell me what's breaking and what it costs you. No preparation needed.
- 2
A written plan and a fixed quote
Within 2–3 days: what I'll build, in what order, and the price. No obligation.
- 3
Build with weekly demos
You see working software every week, not a big reveal at the end.
- 4
Handover and training
Your team learns to run it, with written guides for the people who'll use it daily.
- 5
A care plan
Hosting, backups, fixes and small changes each month, so the system keeps up with the business.
Writing
54 articles on building real systems.
Deep dives from my blog: the same thinking that goes into client work, written down so you can check it.
Latest
- 16 Jul 2026, Industry notesThe X Open Source Gambit: Analyzing Musk's Code Transparency PlayElon Musk's promise to open source X isn't just about code access; it's a strategic maneuver to rebuild trust. We analyze the technical feasibility, the security caveats, and what this means for engineering leaders.
- 16 Jul 2026, Industry notesGLM 5.2 Beats Claude in Our Benchmarks: A Deep DiveOur internal engineering benchmarks reveal a surprising shift: GLM 5.2 outperforms Claude in specific high-load reasoning tasks. Here is the data, the methodology, and what it means for your stack.
- 16 Jul 2026, Industry notesWhy Apple Skipped M6 for M7 Ultra: The 1.5TB Memory BetApple's reported leap to the M7 Ultra with 1.5TB of memory isn't just a spec bump—it's a fundamental shift in how we architect AI applications. Here is why local inference is about to eat the cloud's lunch.
Every article
- 16 Jul 2026The X Open Source Gambit: Analyzing Musk's Code Transparency PlayIndustry notes, 6 min read
- 16 Jul 2026GLM 5.2 Beats Claude in Our Benchmarks: A Deep DiveIndustry notes, 6 min read
- 16 Jul 2026Why Apple Skipped M6 for M7 Ultra: The 1.5TB Memory BetIndustry notes, 6 min read
- 1 Jul 2026Securing the AI Supply Chain: Hidden Markers & Dependency BiasIndustry notes, 6 min read
- 11 Jun 2026The JavaScript Event Loop, Visualized in 3DHands-on guides, 7 min read
- 14 Apr 2026The Virtual Layer of Trust: Why Mongoose Schemas Matter for Scaling StartupsIndustry notes, 5 min read
- 11 Apr 2026Beyond the Toy: Architecting Production Systems with FirebaseBackend and data, 6 min read
- 11 Apr 2026Building Production-Grade Maps with react-native-mapsHands-on guides, 6 min read
Questions clients ask
How much does a project cost?
Defined projects get a fixed price; ongoing or changing work runs on a weekly rate. Either way you get a written plan and quote within 2–3 days of our first call, with no obligation.
How long does it take?
The written plan sets out what gets built, in what order, with dates. You see working software every week, so nothing waits for a big reveal at the end.
What happens after launch?
A monthly care plan covers hosting, backups, monitoring, fixes and small changes, so the system keeps up with your business.
Do you work with clients outside Pakistan?
Yes. I'm in Karachi (UTC+5), with hours that overlap European and US working days. You get a demo every week and quick answers on WhatsApp or Slack in between.
Can you work with what we already have?
Yes. I connect to the tools you already use, like WhatsApp, payment providers, broker APIs and existing databases, or take over an existing codebase. Taking over code starts with a short paid audit, so the quote is honest.
Are you open to full-time roles?
Yes, senior full-stack roles, remote or hybrid. Email me or message me on LinkedIn.