Backend and infrastructure, Stockholm. Swedish citizen, working in Swedish, English and Chinese.
Mathematics at Stockholm University. All degree requirements completed, degree certificate not yet claimed.
Most of my 2026 work is in private repositories, so this page is the part of it that this profile does not show.
Largest contributor to the live branch of a production web platform running on AWS: 558 of 879 commits and around 100 merged pull requests, with review and merge responsibility in a shared repository.
- Owned security remediation. Closed 20+ findings from a pre launch audit, and pinned every fix with its own CI regression test.
- Cut database traffic on the busiest endpoints from 54 to 7 queries per request, then added a per endpoint query ceiling in CI so it cannot silently regress.
- Built the data compliance layer: subject data export, erasure with personal data scrubbing, versioned consent, HMAC pseudonymisation of national identity numbers. Wrote the technical data protection impact assessment.
Python · FastAPI · async SQLAlchemy · PostgreSQL · Celery · Redis · Next.js · TypeScript · Docker · GitHub Actions · AWS EC2 and RDS. Linux throughout.
One of five finalist teams out of 23, organised by 2Hero with Saab Digital Battlespace Solutions and KTH Innovation. A real time browser simulation of an air defence command system, roughly 25,000 lines built in 14 days by a team of five. Largest contributor, with ownership of the reinforcement learning subsystem: a NumPy physics core, a FastAPI WebSocket backend streaming world state at 30 Hz, a Three.js frontend, two custom Gymnasium environments, and a cross evaluation harness that established our rule based baseline matched or beat every trained model we had.
Local conditions and global conclusions in convex geometry (in Swedish). A single proof chain from Radon's lemma through Helly's theorem and the centerpoint theorem to a weak center transversal theorem, kept in affine and convex language so it needs only elementary linear algebra. Closer to computer science than it sounds: Radon gives the VC dimension of halfspaces, Helly underpins linear programming feasibility, and the centerpoint theorem is Tukey depth in robust statistics.
Alongside that I built a forecasting engine that scores its own probability estimates after the fact — Platt scaling and isotonic regression written from scratch, evaluated with Brier score, log loss and expected calibration error over a corpus of 554 resolved predictions.
Ten years of building and repairing computers, as a hobby rather than professionally. Full builds; memory, SSD, drive and graphics card replacement; laptop upgrades; dust cleaning and thermal paste; mouse and keyboard switch replacement, which means soldering; operating system reinstalls across three platforms; basic Linux command line. I follow GPU, VRAM, multi card and AI server hardware closely. This is personal practice, not enterprise data centre experience.
Junior roles in Stockholm, on two tracks: backend and data engineering, and data centre operations, server hardware and IT support. Comfortable with shift work.