Embedded systems → serverless → shipped product
PhD in Computer Science (embedded systems), postdoctoral research in serverless computing, and hands-on work with state-of-the-art AI/ML. I design and ship websites, AI assistants, and mobile apps for teams that want research-grade engineering.
Each engagement starts with a fixed scope, a timeline, and a quote — so you know exactly what you are getting before anything is built.
Marketing sites, landing pages, and full web apps — fast, accessible, SEO-ready, and deployed on modern infrastructure with analytics wired in.
Next.js · React · TypeScript · Tailwind · Postgres · Vercel / Cloudflare Fixed-scope quotes · retainers availableAssistants trained on your documents, with retrieval, tool use, evaluation, and human hand-off — from a website widget to an internal agent.
RAG · vector search · function calling · eval harnesses · Anthropic / OpenAI APIs Fixed-scope quotes + monthly hosting & iterationCross-platform iOS and Android apps from MVP to store launch, with clean API integration, offline support, and push notifications.
React Native · Expo · Flutter · REST / GraphQL · SQLite Fixed-price MVP · phased deliveryApplied research for teams working at the edge: embedded ML, energy-aware systems, serverless architecture, and ML-systems design. Prototyping, benchmarking, and co-authorship.
Embedded ML · edge AI · serverless cost / latency modelling · systems benchmarking $50/hour · grant-funded · equity consideredA cross-section of the kinds of problems I take on — from firmware to cloud to front end.
Devices post batches of sensor readings; the API stores them, checks alert rules and notifies once per violation, even when a device retries. Designed for at-least-once delivery: idempotent writes, a dead-letter queue, least-privilege IAM asserted in tests, and per-device keys that are never stored. 119 automated tests, including the DynamoDB adapters run against an emulator. Infrastructure is verified by synthesis, not yet deployed.
A portable C core behind a hardware abstraction layer: ring-buffered readings, exponential backoff with jitter, correct timestamps even before the clock is set, and an integer anomaly detector. About 3.8 KB of code on Cortex-M. 11,163 checks pass, including under the undefined-behaviour sanitiser, and the real firmware is tested end-to-end against the real API handlers through outages and lost responses. Not yet run on physical hardware.
A retrieval-augmented assistant that streams answers with citations, shows the passages and scores it used, and can open a ticket or hand off to a person. Runs offline with no API key (extractive mode) or against Claude. An evaluation harness scores retrieval, answers and tool routing, currently 10 of 10 on its test set.
Log expenses, track a monthly budget, see where the money goes, export to CSV. Everything lives on the device. Money is stored as integer cents, input parsing is deliberately strict, and budget projections count rent once instead of scaling it. 26 tests against real SQL; bundles for iOS and Android and exercised end-to-end in a browser, but not yet run on a phone.
A four-page, content-driven site for a local business: services and pricing, team, and a contact form with validation and spam protection. Accessible, with per-page metadata, a sitemap and structured data. Re-skinning it for a new client means editing a handful of content files.
From the firmware layer to the interface — grouped the way the system actually stacks up.
My doctoral work was in embedded systems — my thesis, Coordinated concurrent memory accesses on a reconfigurable multimedia processor, showed how a mobile multimedia accelerator can use many local memories in parallel, through a shared-memory programming model and automated, deterministic scheduling of memory accesses. My postdoctoral research focused on serverless computing — at the University of Manchester I helped design a heterogeneous infrastructure that combined VMs, micro-VMs, unikernels and scalable storage, and co-authored a paper on unikernel-based edge computing for 6G networks. I keep current with state-of-the-art AI/ML and actively want research projects to contribute to, alongside client work.
A call to pin down the goal, constraints, and the one metric that defines success. You get a written summary.
Fixed scope, timeline, milestones, and price. Revisions and change-handling agreed up front.
Weekly demos against the plan. You always see progress you can actually use.
Deploy, QA, analytics, documentation, and a handover walkthrough so your team can run it.
An optional retainer for updates, monitoring, and iteration once you are live.
A defined deliverable with milestone payments: a deposit to start, the rest across agreed milestones.
A block of hours each month for iteration, fixes, monitoring, and small features. Cancel with 30 days' notice.
Advisory, prototyping, benchmarking, and co-authorship. Scoped by the week or by deliverable.
For early-stage teams with a clear vision — a lower cash rate in exchange for a modest equity stake.
My rate is $50/hour. Every project still gets a fixed quote after discovery, built from that rate, so there are no hourly surprises.
I'm Samar — a computer scientist and software engineer. I hold a PhD in computer science specialising in embedded systems and completed postdoctoral research in serverless computing, and I've spent more than 15 years building software that ships. I work with founders, small businesses, and research teams who want the discipline of research applied to real products.
Based in Karachi, Pakistan · experience across France, the UK and Pakistan · working with clients worldwide.
A project, an app, an assistant, or a research problem you're stuck on — send a few lines about it and I'll reply with next steps.