Computer Science · ETH Zurich

Hi, I'm Hafsa.
I build apps people use, and I like making software fast.

I'm a master's student in Computer Science at ETH Zurich, focusing on data management systems and software engineering. I design, build and ship my own apps from the first sketch to the app store.

About

I'm based in Zurich and study Computer Science at ETH Zurich, where I also did my bachelor's. My master's major is Data Management Systems, with a minor in Programming Languages & Software Engineering.

I like understanding the whole stack, from the hardware and systems underneath up to the software people actually use. On the software side, I enjoy the whole path of a product: working out the data model, building the backend, writing the tests, setting up CI, and getting the app into people's hands. I have worked on multiple projects that are live or about to launch. On the systems side, I've vectorised kernels by hand, scheduled workloads in the cloud and tested hardware interconnects for security flaws, because I want to know what the machine is actually doing.

Since 2024 I've been a teaching assistant for Digital Design and Computer Architecture, helping students through their first steps with Verilog, MIPS assembly and FPGAs.

Based in
Zurich, Switzerland
Studying
MSc Computer Science, ETH Zurich
Focus
Full stack, from hardware and systems to software
Languages
German (native), English (fluent), French (basic)

Projects

Real or AI? banner: ten images a day, can you spot the AI ones?

Real or AI?

Ten images a day. Can you spot the AI ones?

A web game that tests whether you can tell real photos from AI-generated ones. Blitz mode gives you 5 seconds per image and 3 lives, and a daily challenge of 10 images is the same for everyone, with a score to share.

  • A nightly pipeline takes public-domain photos and generates a matching AI "twin" for each with a vision model and FLUX, then normalises both (1024×1024, metadata stripped) to hide the label
  • Serverless API on Cloudflare Workers (Hono) with a D1 database and R2 image storage
  • Security built in: rate limits, hashed one-time sign-in codes, AI-moderated comments and a strict Content-Security-Policy
  • ~120 unit and API tests plus Playwright end-to-end tests, with continuous deployment through GitHub Actions
  • TypeScript
  • Astro
  • Svelte
  • Hono
  • Cloudflare Workers
  • Workers AI
  • Playwright
realoraiquiz.com
KnowHouse banner: share the chores, keep the know-how

KnowHouse

Share the chores. Keep the know-how.

A family app for household tasks: assign and accept to-dos, earn points, climb the leaderboard and keep instructions for how things are done at home. It started as a team project and is now published on Google Play.

  • Real-time collaboration, shared calendar, personalised task recommendations and user profiles
  • 200+ unit, provider and widget tests
  • CI/CD with GitHub Actions: every release is versioned, signed and uploaded to Google Play automatically
  • Flutter
  • Dart
  • Firebase
  • Firestore
  • GitHub Actions
getknowhouse.com
Shax banner: the Somali board game

Shax

The traditional Somali board game.

A digital version of Shax, a two-player strategy game related to Nine Men's Morris. Play offline against the computer or a friend on the same device, in English, German or Somali.

  • Minimax game AI with three difficulty levels
  • The full traditional rules, including the removal phase and the Oodan rule
  • One Flutter codebase for Android, iOS and the web, with automated signed releases
  • Flutter
  • Dart
  • Game AI
  • i18n
  • GitHub Actions
shax.app

University work

FastGATr

1500× faster than naive C++ · 1.7–4.5× faster than the optimised PyTorch reference

Optimised a geometric algebra transformer on Apple M1. Exploited structure-tensor sparsity with generated kernels (about 140× fewer operations), added ILP, operator fusion and cache-aware weight repacking, and hand-vectorised the hot loops with NEON intrinsics.

  • C++
  • Python
  • ARM NEON SIMD

Cloud workload scheduling

Latency-sensitive and batch jobs sharing one cluster · team of 3

Built schedulers that co-locate memcached with PARSEC batch workloads. Measured resource interference with iBench and PARSEC scalability, then used the results to schedule on a heterogeneous VM cluster (static load) and on a single machine (dynamic load). We then evolved our hand-written policy with OpenEvolve and Llama 3.3 70B over 100 iterations, using our own scoring function (makespan, SLO risk, invalid schedules) and system prompts tuned so the LLM would try running jobs in parallel. The evolved policy cut the total runtime by 19% with zero SLO violations.

  • Google Cloud
  • Kubernetes
  • Kops
  • Python
  • OpenEvolve
  • LLMs
  • Prompt engineering

Compiler Design Project

Oat → LLVM IR → x86-64, with dataflow optimisations and register allocation

Built a compiler for Oat, an object-oriented language, in OCaml. Step by step: an x86-64 simulator and assembler, a backend that compiles LLVM IR to x86-64 assembly, a frontend with lexing, parsing and type checking that lowers Oat to LLVM IR, and finally dataflow-analysis optimisations and register allocation.

  • OCaml
  • LLVM IR
  • x86-64 Assembly

Bachelor Thesis: XRAY+

300,000+ automated test cases on AXI interconnects

Written at ETH Zurich's Secure and Trustworthy Systems Group. Extended XRAY, a tool for security analysis of AXI interconnects, with new execution modes and security properties, which uncovered previously unknown security vulnerabilities. Automated large-scale testing with a TCL pipeline script running over 300,000 test cases on different AXI interconnects.

  • Verilog
  • SystemVerilog
  • Vivado
  • TCL

Smaller projects

Privacy Tracker

What websites learn from your permissions

An interactive web app that builds the data profile a site could infer from permissions like location, camera and contacts, and how it grows the longer you're tracked. Personas show realistic inferences and the ads they lead to, in English and German. 430+ tests; every push is tested and deployed with GitHub Actions.

  • Next.js
  • React
  • Tailwind CSS
  • Jest
  • GitHub Actions

Zürcher Taxi

A trilingual landing page for a Zurich taxi driver

A single-page site in German, English and Arabic, with a right-to-left layout for Arabic from the same stylesheet and WhatsApp booking in the visitor's language. Plain HTML, CSS and JavaScript with no build step, deployed to GitHub Pages on every push.

  • HTML
  • CSS
  • JavaScript
  • i18n / RTL
  • GitHub Pages

Experience & education

  1. 2025 – 2028

    MSc Computer Science · ETH Zurich

    Major in Data Management Systems, minor in Programming Languages & Software Engineering. Courses include Big Data, Cloud Computing Architecture, Advanced Systems Lab, Concepts of Object-Oriented Programming and AI in Education.

  2. 2024 – 2026 (spring)

    Teaching Assistant · ETH Zurich

    Digital Design and Computer Architecture. Guided students through 10 hands-on lab sessions on digital circuit design, Verilog, MIPS assembly, FPGAs and system integration, and reviewed their lab submissions with feedback.

  3. 2022 – 2025

    BSc Computer Science · ETH Zurich

    GPA 5.15 / 6. Bachelor's thesis at the Secure and Trustworthy Systems Group (6 / 6). Courses include Parallel Programming, Systems Programming, Operating Systems, Computer Networks, Compiler Design, Data Modeling and Databases, Information Retrieval and Machine Learning.

Skills

Languages

  • Java
  • Dart
  • C
  • C++
  • Python
  • JavaScript
  • TypeScript
  • SQL
  • OCaml
  • Haskell
  • Verilog
  • SystemVerilog
  • TCL
  • x86-64 Assembly
  • HTML/CSS

Frameworks & tools

  • Flutter
  • Firebase
  • React
  • Svelte
  • Astro
  • PostgreSQL
  • PyTorch
  • Cloudflare Workers
  • Kubernetes
  • Google Cloud
  • GitHub Actions
  • Git
  • Vivado
  • OpenEvolve
  • LLMs

Contact

Whether it's about a role, a project or one of my apps, I'd be glad to hear from you.