Real outcomes, verifiable metrics
Production cases with names anonymized at clients' request. Each one has documented benchmarks, scope, and timeline — available under NDA when required.
Verifiable metrics
A European retail SaaS's reporting module got slower the wider the range (month → quarter → semester), and the latency was bleeding into inserts and internal services. UUID v4→v7 migration across distributed databases with billions of records (custom Rust tool, Jaeger tracing). A full Go rewrite avoided.
Gastroenterology clinic: a self-hosted conversational AI (Llama + LangChain) with answers restricted to clinical guidelines, over app and WhatsApp. No per-message fees, data stays in-house (LGPD-friendly).
Solar-energy company: a full CRM plus a real-time pricing engine (cost + tax + margin). Proposal generation dropped from 2–3 days to real-time.
ESG-tech startup with on-chain audit trails: monorepo refactor (pnpm + Nx), Rust/Substrate pallets for the certification lifecycle, and a Kubernetes (EKS) deploy. ~50% less dev-environment memory.
Outcomes in production
Each case below is a real problem, taken to production, with a verifiable metric. These are not prototypes or proofs of concept — they are systems that kept running after we delivered.
+260% write performance — retail-SaaS reporting module
Sector: European retail SaaS. Tag: Performance.
The financial and sales-reporting module (Node.js + Express + TypeORM) of a European retail SaaS was degrading badly: the larger the report range — month, quarter, semester — the slower it got. The latency started bleeding into inserts and other internal services, and the company was about to rewrite the whole subsystem in Go.
The diagnosis showed the problem was not the language. Profiling and benchmarking pinpointed the bottleneck in the database, tied to the use of random UUID v4 as the key. The fix was surgical, not a rewrite: we migrated the keys from UUID v4 to v7 (time-ordered) across 6 distributed databases processing billions of records, using a custom migration tool written in Rust and instrumented with Jaeger tracing. Result: +260% write performance (with proportionally larger gains on reads and joins), a major reduction in infrastructure cost, and the costly Go rewrite avoided.
Self-hosted clinical AI — gastroenterology clinic
Sector: healthcare / medical clinic. Tag: AI Chatbot.
A gastroenterology clinic seeing roughly 380 patients per month needed to reduce no-shows, increase exam completion, and ease front-desk load — without patient data leaving its own infrastructure.
We built a patient-journey system — scheduling, digital anamnesis, and pre/post-procedure guidance — with an on-premise, self-hosted conversational AI (Llama + LangChain) whose answers are restricted to clinical guidelines: the AI is grounded and will not answer outside its sources. It is delivered through a mobile app and WhatsApp, with self-service kiosks at the front desk. No per-message platform fees, and data stays in-house (LGPD-friendly). Results: +25% appointment attendance, +40% endoscopy exam completion, and reduced front-desk load via self-service.
+280% revenue — solar-energy CRM and real-time pricing
Sector: solar energy (Brazil). Tag: AI Automation / Development.
A Brazilian solar-energy company had a slow sales cycle and manual pricing. We built a full CRM — leads, proposals, contracts, and sales funnel — and an intelligent real-time pricing engine (cost + tax + margin), with weekly batch sync of hundreds of SKUs across distributors and ERP invoice integration.
Results: +280% revenue over 12 months, lead conversion down from 30 days to 7–10 days, and proposal generation cut from 2–3 days to real-time.
2–3× faster builds, −50% dev memory — ESG audit platform
Sector: ESG-tech / blockchain. Tag: Architecture / Rust.
An ESG-tech startup was building a corporate audit and certification platform with cryptographically signed, on-chain audit trails. We restructured the monorepo (pnpm workspaces + Nx, 4 projects in parallel) and added full unit, integration, and E2E test coverage.
As the team's Rust engineer, I wrote the custom Substrate (Polkadot SDK) pallets modeling the full audit → validation → on-chain-certificate lifecycle, integrated with NestJS services and deployed on AWS EKS. Results: 2–3× build and runtime performance, and roughly 50% less dev-environment memory.
Got a similar case?
Critical performance, AI in production, modernization, or a senior squad — describe the scenario at /en/contact and we respond with feasibility within 24 business hours.