Coderland.

Invoice-factoring platform with a deployment per country and portals for administrators, clients and affiliates, integrated with a partner fintech API for funding.

• Root-caused a silent integration outage: after a storage migration, signed file URLs exceeded the partner API's 500-character limit and every funding submission was rejected for 11 days. Built a permanent short-URL proxy that signs links on demand, keeping storage private and auditable, and recovered 66 of 67 affected deals.
• Designed and built a digital-consent evidence module: an append-only record of IP, server timestamp, user agent, the full accepted text and a SHA-256 hash, plus PDF certificates. Covered by 48 tests and live in all three countries.
• Led the move of document storage to cloud storage (1,627 files) and a 101-commit country release with 16 migrations, resolving six unplanned blockers along the way.
• Fixed 504 timeouts on large PDF reports: profiling showed rendering, not SQL, was the bottleneck, so generation moved to a dedicated background queue that also releases memory after each job.
• Traced recurring 504 outages to outbound HTTP calls without timeouts hanging the application workers, and shipped timeouts plus database indexes.
• Found a float × Decimal bug that had been silently dropping batch uploads in two countries since 2024, and fixed a portal-wide 403 for read-only users across 32 templates with a test that walks every link.

Legacy Java EE system managing catalog, pricing, orders and marketplace connections for a high-volume retailer.

• Root-caused a payment-integrity defect: a background job confirmed marketplace payments as successful for every order, including orders that did not exist, and contained it.
• Restored store access to online orders for 176 users by tracing a missing permission group in the new per-store logins — fixed in data, with no code change.
• Recovered 616 of 616 seller invoices for the finance team after a seller suspension by scripting the right marketplace API, with no new development.
• Authored a production AWS infrastructure audit that identified roughly $12.5k per year in potential savings.
• Wrote the production cutover runbook for the catalog integration with the VTEX commerce platform.

Reactive service connecting the public careers site and the employee portal with a recruiting SaaS: vacancies, applications, referrals and webhooks.

• Built an AI job-posting pipeline: a poller and an outbox worker turn draft vacancies into the company's standard template with Gemini, later extended to postings published directly.
• Automated a candidate re-engagement campaign with signed-token links, a durable outbox with dead-letter handling and email de-duplication, replacing manual spreadsheet cleanup.
• Root-caused reopened and duplicated job postings to an external CRM automation after first ruling out the middleware.
• Moved the service to a fully non-blocking WebFlux stack on Java 25 / Spring Boot 4, with client timeouts, connection pools and Prometheus monitoring.
• A post-upgrade audit uncovered a CI pipeline silently skipping all 20 test classes and a blocking call on the reactive event loop — both fixed.

Portal used for timesheets, invoicing, leave requests and referral commissions, integrated with the CRM, issue trackers and the recruiting middleware.

• Modernized the full stack with a strangler approach, one library per small PR: Spring Boot 3.5 / Java 17 on the backend and Vite, React 19 and TypeScript on the frontend.
• Replaced a CRM SDK integration spread across 168 files with lean REST calls, and resolved duplicate CRM contacts deterministically, returning an explicit support error code instead of guessing.
• Investigated referral commissions on production data: only 2 of 310 referrals were ever hired, showing an attribution and conversion problem rather than data loss, and fixed a double-charge root cause.
• Root-caused around 40 recurring payroll and billing incidents, such as an API page limit truncating results and cancelled pre-invoices counted as pending.
• Switched the build to pnpm, cutting CI build time from ~8 minutes to ~3 minutes.

Multilingual corporate website and headless CMS whose contact forms feed leads into the CRM.

• Hardened lead forms with four anti-spam layers plus a final AI check with Gemini, after an audit found the reCAPTCHA token was never verified on the server.
• Traced a long-running silent failure in lead capture through the logs and restored it, then fixed lead-source attribution so the CRM stores the real UTM source instead of a single default.
• Migrated the blog off a paid SaaS into the CMS: 67 bilingual articles with permanent redirects and webhook-driven cache refresh.
• Brought 45 orphan URLs to 100% crawlable coverage and fixed redirect status codes, validated in production.
• Wrote a blameless postmortem for a cache-poisoning incident, identifying a three-layer amplification, and shipped the hotfix and architecture fix; also led the Strapi 4 → 5 production upgrade.