Most AI demos look great and then break on real data and real users. I build the backend, the app and the model logic around the way your business actually works.
133 projects done. 71% were for clients who came back. Top Rated on Upwork, 100% Job Success, 106 five-star reviews.
I add AI to the product you already have. It talks to your database, your CRM and your APIs, and it does real work, not demos.
A chatbot that answers from your own documents. Every answer shows where it came from, so you can check it.
Agents that do the boring steps: sort leads, update the CRM, send the email, book the slot. A person approves before anything important goes out.
The whole app, by one engineer. Database, API, screens, logins, payments and deployment. React, Next.js, Django, Node and Supabase.
Read text and fields out of photos, PDFs and ID documents. Find things in images. Send the unclear ones to a person instead of guessing.
An old PHP or Yii2 system that nobody wants to touch. I make it stable, add what you need, and leave notes the next person can follow.
A few builds that show how I work: understand the job first, ship it, then check it does what the business needed.
Sensor-based AI system for live handoff detection, confidence scoring, and owner/non-owner validation with immediate authorization control.
View flagship case study
MPhil research application: facial-expression analysis in Python, OpenCV, TensorFlow and scikit-learn to estimate depression and anxiety levels. University of Karachi, 2019 to 2021.
MPhil research project, not a client engagement
Android app with a C++ DSP core on Oboe: noise suppression, feedback control and a limiter, built for low latency.
View case study
The same purchase posts the same journal entry in Odoo 18, Zoho Books and SAP Business One. The accounting is tested once.
View case study
Duplicate and missed onboarding emails after payment. A two-day audit as its own milestone with written findings, then a second contract for the fix. January 2026.
View case study
A 2023 GPT-3 proof of concept: chunk, embed, retrieve by similarity, answer from the retrieved passages. The pattern now called RAG, built in plain Python.
View case study
Customer app, driver app (React Native), React admin for restaurants and a Django REST backend with Stripe Connect payouts and PostGIS. Six milestones, 2020 to 2021.
Read case study ↗
Laravel + React web and two React Native clients. A healthcare product modernization with patient booking, doctor workflows, shared APIs and security regression tests.
Read case study ↗
Django, Celery and React SaaS with FastAPI engines for face search, AI image generation, LoRA training and virtual try-on. GPU jobs are queued.
Read case study ↗
My own SaaS: finds and qualifies businesses, drafts outreach a person approves, runs an AI receptionist and learns which proposals win. Next.js, Supabase, BullMQ.
Read case study ↗
Five agents with one job each: scout, writer, reviewer, approver, editor. The approval step is plain code, so nothing goes out without a tap on the phone. Claude, Codex, MCP.
Read case study ↗
Partner referral and rewards platform, partners share a link, track each student's journey, and choose an impact reward. React 18 + TypeScript on Django 5 / DRF with PostgreSQL.
Read case study ↗
Phone-sensor behavioral authentication PoC, on-device ML (scikit-learn), FastAPI backend, and a security-first owner/handoff state machine, validated with a cross-day signal feasibility study.
Read case study ↗
Medicine-donation marketplace on Expo (React Native) and Supabase, iOS, Android and a real desktop web app, bilingual EN/UR, with the donor's address and handover code protected by database policies rather than hidden UI.
Read case study ↗
Seven Lambdas, two Deluge scripts and a 2.9 GB directory rebuilt as one Odoo 17 module, run against the client's own files. Nine defects found in the original.
Read case study ↗
Built at Baydot. Pakistan stock portfolio during market hours, RSI notices on your phone, and a five-year test of the rule before any alert was built. FastAPI, SQLite, 345 tests.
Read case study ↗
A doctor and patient app pair on a Laravel API, built and extended across three contracts from 2018 to 2020.
Read case study ↗
Android live mic-to-speaker app with a custom C++ DSP core, noise suppression, feedback/howling suppression, and a look-ahead limiter on Oboe low-latency audio.
Read case study ↗
ComfyUI graphs that keep one face consistent with a LoRA, IP-Adapter FaceID Plus V2 and an OpenPose ControlNet, plus the kohya_ss training notebook.
Read case study ↗
An investment ledger whose accounting lives in one package: the same purchase posts the same journal entry in Odoo 18, Zoho Books and SAP Business One.
Read case study ↗
Android MediaCodec transcoding and compression, then native overlay and baking on iOS and Android, exposed to the app as typed React Native modules. Two contracts for one client, 2025 and 2026.
Read case study ↗
A February 2023 proof of concept: chunk, embed, retrieve by similarity, answer from the retrieved passages. The pattern now called RAG, in plain Python and the OpenAI API.
Read case study ↗
Duplicate and missed onboarding emails after payment. A two-day audit as its own milestone, written findings, then a second contract for the fix. January 2026.
Read case study ↗
FFmpeg encoding on AWS Lambda with an auto-scaling EC2 overflow tier, six contracts with the same client between 2016 and 2018.
Read case study ↗
Google Vision and Document AI extraction for a document scanning app, plus passport and licence OCR behind a React Native backend.
Read case study ↗
21 product contracts for one platform company that builds on Django and React, between 2019 and 2025, parallel builds on shared standards.
Read case study ↗
Augmented reality app, live on iOS and Android since January 2025. I built the whole app: maps, AR, location-based outdoor experiences and leaderboards, plus a Django admin where staff pick regions on a Mapbox map.
Visit live site ↗
Ask questions about a large claims spreadsheet in plain English. Built in Next.js + Supabase, rewritten in Go: read-only SQL guard, 400,000 rows loaded in 18 seconds.
Read case study ↗
Durable jobs for long image generations: leases, checkpoints, retries and backpressure. A crashed worker's job resumes without a second GPU run.
Read case study ↗
Paste a job post or screenshot: matched CV, honest cover letter, sent from Gmail only after you tap Send.
Read case study ↗
Texting inside a voice AI recruiting product: queued sends, delivery status, no duplicates, voice session race fixes. Next.js, Supabase, BullMQ.
Read case study ↗
iOS coloring app with scanline flood-fill rendering, tap-to-color interaction, and a curated palette experience.
Read case study ↗
React Native app with ID-scan sign-up, covers and photo claims on a Cognito, Lambda and DynamoDB backend. 2018 to 2019.
Read case study ↗
Android messenger with phone login, chat, Sinch voice calls, push and an admin app. 2016.
Read case study ↗
Lambda job that diffs about 28 public policy pages and emails every change for review.
Read case study ↗
Record, trim with FFmpeg and upload to YouTube, plus an admin app and an AngularJS web app.
Read case study ↗Public feedback from completed Upwork jobs, backed by 100% Job Success and Top Rated status.
71% of all my contracts came from repeat clients. 18 clients hired me two or more times, one of them 21 separate times, across a relationship that ran from 2019 to 2025. Clients come back because the work holds up in production.
Choose from eight focused role resumes, each with matching project evidence, a single-column PDF and an editable text version. Forward Deployed AI Engineering is my primary focus.
Take these as a starting point. On the call we agree what you actually need and how long it takes.
I understand your business goal, current system, target users, budget, and launch timeline.
You get a plain plan: what gets built, in what order, on which stack, and how long it takes.
I ship in short cycles. You see something working every week and can change direction early.
I deploy it, hand it over with notes, fix what breaks, and stay for the next step.
Tell me what is broken or what you want built, and I will tell you what I would do first and how long it takes.
