Selected work
A few of the platforms we've built and kept running — chosen for what they show about how we work under real conditions: high traffic, sensitive data and systems that aren't allowed to go down.
Client names are withheld under NDA. Numbers and technical detail below are accurate; identities are described by category only where required. For serious inquiries, we can also prepare a depersonalized demo environment on request.
A pan-European public institution promoting linguistic diversity
A 43-language platform that holds up under a near-1M-visit day
A content-heavy site serving interactive tools, event listings, games and educational material in 43 languages — with traffic that climbs through the final week of the campaign and peaks sharply on a single key date each year, growing year over year.
43languages served
~1Mvisits on peak day
4.3M+visits across the campaign month
We handle the multilingual content architecture and delivery, search and indexing across the full language set and the database and infrastructure work that keeps response times stable through that week-long ramp-up and its peak — not a one-off stress test, but a pattern the platform meets reliably every year.
A national government identity-verification authority
An identity API trusted by banks and employers, running at 500+ requests a second
A high-security API that lets banks, employers and other authorized organizations verify a citizen's identity against the national registry in real time, against a database holding over 750 million historical records.
500+requests per second, sustained peak
2.4M+successful verifications a month
750M+historical records in the archive
We migrated a legacy platform to a standalone modern API without a second of downtime — including building a purpose-made proxy to take over the highest-traffic endpoint during the old server's upgrade window, so the cutover carried zero risk to live traffic. On the database side, we traced a recurring performance bottleneck to a quota check running over 100,000 times an hour, replaced it with a pre-aggregated counter and cut response times by roughly a quarter.
Built-in resilience, not just monitoring. When an hourly usage counter silently stopped updating correctly under peak load, we didn't just fix the bug — we built an automated consistency-check-and-repair service that catches and self-heals that class of failure going forward.
A government identity-document service
39K+ appointment bookings in the first five days, zero downtime
A public booking platform for citizens to schedule in-person appointments at government offices, with real-time identity verification, OTP confirmation, capacity-limited time slots and multi-channel notifications — plus an embeddable widget partner organizations can host on their own sites.
39K+bookings in the first 5 days live
10K+bookings in a single day at peak
100%uptime through launch week
We built the full architecture — a React front end, a .NET API, Azure SQL, secrets managed in Key Vault — engineered for a target of 30,000 registrations a day. Every deployment to production passes through a mandatory human approval step before going live.
Engineering under the hood. Every booking runs a real-time identity check against an external government system inside a single transaction — with lockout protection, duplicate-request guarding and a timing-attack-resistant comparison on sensitive data — then locks the time slot against double-booking under concurrent load, while notification delivery is kept outside that transaction so a slow SMS provider can never hold up a booking.
A large enterprise client
Search functionality rebuilt for guaranteed reliability
A public-facing content and publication platform for enterprise documents — with multilingual full-text search across over 100,000 published documents, spanning years of historical records.
100K+documents indexed and searchable
8content categories covered
7languages searched simultaneously
We rebuilt the search-indexing pipeline into a durable, queue-based system designed for guaranteed delivery rather than raw speed — with cross-language search built in from the ground up, so a single query is matched correctly across every supported language rather than just the one it was typed in, including languages with fundamentally different structures like Arabic and Chinese. It runs across a genuine cross-cloud setup, with pipelines orchestrating deployment to cloud-hosted infrastructure.
We don't just fix bugs, we build the tooling to prove they're fixed. When documents began silently disappearing from search results — some dating back years — we ran a methodical, months-long investigation and found four distinct root causes, including a subtle memory-pressure issue that caused writes to silently drop under sustained load. We fixed the underlying cause, then built a purpose-made audit tool that checks every record against the index to confirm nothing is missing.