I build backends that scale under load.

New scenarios create new needs. The 2020 pandemic changed how people interact with brands, and live commerce — the old TV shopping channel reimagined for the web — became how fashion retail reached customers who could no longer walk into a store.

Grupo SOMA became a pioneer of the format in Brazil. The platform brings streaming, live chat and checkout into a single flow, so a viewer can watch a presenter show a piece and buy it without leaving the broadcast.

I worked on it as part of the full-stack team Digital Republic embedded with the client.

Original case study ↗

Lojix is Farm's first fully interactive, fully digital store. It was built for customers who want the convenience of e-commerce but not the coldness of it — the format keeps the human part of walking into a Farm store.

Instead of conventional salespeople, the brand's own team presents the pieces: stylists, print designers, fashion production and communication staff show the collection live, answer questions in chat, and offer pieces with on-the-spot discounts.

It sits inside Grupo SOMA's wider live commerce push, and was one of the earliest examples of the format working at scale in Brazilian fashion retail.

Original case study ↗

Farm na Nuvem was Farm's first carbon-neutral store, and it had no floor: a photographic 360° tour of the real shop that you navigate the way you would walk it, from the planted facade on the street to the racks inside.

The point was that it is not a showroom. Every piece in the tour is a hotspot — open the gallery, filter by category, land on a product with real pricing, discounts, instalments, size availability and live stock status, then add it to your bag without ever leaving the panorama.

That means the 360 layer sits on top of a working commerce backend: catalogue, stock and checkout all have to answer in real time, or the illusion of walking through a store collapses into a broken picture.

Grupo SOMA is a Brazilian luxury fashion group formed in 2010 from the merger of Farm and Animale, and it now holds many of the country's leading fashion brands. That portfolio runs on a lot of web surface area.

I was embedded in the client's web squad rather than working at arm's length — engineers were picked for technical depth and fit with the existing team. The work centred on evolving the group's web and e-commerce applications and making them measurably faster.

Beyond shipping features, much of the value was raising the digital maturity of the in-house team, so the improvements outlived the engagement.

Original case study ↗

Stock Car Pro Series is the biggest motorsport championship in Brazil, and a race there is decided in fractions of a second. Watch one and the obvious thing is speed. The non-obvious problem sits in race control: a full grid, all of it happening at once, and every incident has to be seen, judged and recorded while it is still happening.

Get that wrong and the sport gets it wrong — a missed flag or a late call changes the result of a race that thousands of people are watching live.

The Race Control System is what makes that practical: reporting occurrences straight from race control, checking alerts, and surfacing incident data fast enough to act on rather than review afterwards. Decisions land on evidence that arrived in time.

Digital Republic's scope covered the software itself, evaluation, and supporting the rollout and live race-day operation.

Original case study ↗

Finding the right person to talk to is the bottleneck in most careers — the mentor who unlocks an insight, or pressure-tests a business idea, is rarely someone you already know.

Bluementor removes the geography from that problem: find, schedule and talk to mentors entirely online, wherever you are. The product centres on 1:1 meetings with specialists.

The company's ambition was to become the largest career and business mentoring platform in the Nordic region. Digital Republic delivered it end to end.

Original case study ↗

FunderPro is a prop trading firm: traders prove themselves on a challenge account, get funded, and trade the firm's capital with rewards paid out in hours rather than weeks. Everything behind that — accounts, rules, payouts — is backend work where being wrong costs real money.

The platform is white-label: many brands run on one codebase. I work mainly on the money paths. I built wallet checkout with Apple Pay and Google Pay, retrying across payment providers when one fails, and the trader payout flows through bank and crypto providers.

Payouts need guarding, so I built the fraud and risk controls around them: risk checks, duplicate-account and duplicate payout-address detection backed by device fingerprinting, and KYC expiry and name-mismatch checks before money leaves.

I also replaced a third-party loyalty integration with an in-house points program (earn on paid purchases, redeem at checkout), configurable per brand with feature flags, and cut vendor and infrastructure costs, swapping a paid translation service for a Claude API pipeline covering 8 languages and trimming log volume and error-tracking noise.

On the codebase itself, I led the TypeScript strict-mode migration across the NestJS monorepo (strictNullChecks plus four more compiler rules) and hardened message-queue reconnects and scheduled-job error handling.

Original case study ↗

StopChurn builds post-sale experiences that feel personal but run automatically, turning a company's existing customer database into a retention engine instead of a dormant list.

I joined a legacy system with critical reliability issues and very little documentation, and the first job was making it stable enough to build on. From there I delivered features across both backend and frontend while fixing production issues in flight.

The lasting change was introducing automated testing — unit and E2E — which cut regressions and made deploys something the team could trust. It was a low-structure environment that ran on high individual ownership.

Original case study ↗

Variable compensation is one of those processes that stays on spreadsheets far longer than it should. The rules get intricate, the structures get sophisticated, and every month someone recalculates by hand and hopes they got it right.

SplitC models any calculation as a visual flow: data sources feed calculation nodes, which feed payouts and dashboards. It pulls from any system with an API, a database, or even a spreadsheet export, and integrates with the major CRM and ERP platforms.

The platform also collects, validates and approves invoices in place, running compliance checks automatically — matching invoice values against payments, verifying tax IDs, flagging what fails before it becomes a problem.

As a senior backend engineer I migrated the commission engine's heavy workloads from MySQL to BigQuery, cutting processing time from over 30 minutes to under 5, and built event-driven processing on Pub/Sub. I then became tech lead of the squad responsible for performance and scalability.

Original case study ↗