I build backend and data systems for places where a wrong answer is expensive: money that must balance, health records that must stay private, and regulations that must be enforced in code rather than in a policy PDF.
My method is the same in every repository. I name the invariant, find the failure that would break it, prove it holds with a test against a real database, then measure the cost and write down the trade-off.
|
ledger-recon A double-entry ledger with exactly-once posting and end-of-day reconciliation against card, EFT/FAST and MT940 files.
|
medula-precheck Healthcare billing rules as versioned, effective-dated code, with a set-based PL/SQL bulk path that matches the .NET engine finding for finding.
|
|
kvkk-data-inventory Finds, classifies and masks personal data across two database engines. Masking is keyed and one-way, and preserves checksums.
|
dotnet-service-template The baseline every service starts from, installed with
|
Every number above comes from a reproducible run. Each repository's docs/benchmarks.md records the hardware, the host load and the method.
Invariants live in the database. Deferred constraint triggers, append-only history and least-privilege roles. Application code is a second line of defence, not the first.
Tests attack, they don't confirm. Property-based tests, concurrency races and golden cases all run against real PostgreSQL and Oracle through Testcontainers.
Numbers or it didn't happen. A performance claim ships with its benchmark, its environment and its outliers.
Decisions are written. Every repository has ADRs listing the options I rejected and why, plus an honest Limitations section.
Privacy is a design input. Before a schema exists, I decide who can see the data, how long it lives and what trace an access leaves.
Currently building
| Area | Repositories |
|---|---|
| AI infrastructure | guarded-db-mcp, a read-only MCP server for Oracle and PostgreSQL with PII masking and audit · rag-tr · tr-rag-eval · oracle23ai-rag |
| Distributed systems | offline-pos-sync · webhook-relay · idempotent-api · order-saga-temporal · cdc-pipeline · outbox-lab |
| Healthcare & privacy | patient-feedback-workflow · hbys-fhir-bridge · sut-rules · kvkk-dsar · field-crypto · merkle-audit |
| Databases | ora2pg-verify · saas-rls-starter · plan-guard |
| Finance | fx-bitemporal · bank-statement-recon · trkarekod · ubltr-validator |
| Reliability | slo-chaos-lab · flags-rollout · signalr-scale-lab |
Toolbox
| Languages | C# · SQL · PL/SQL · Python · TypeScript |
| Backend | .NET 10, ASP.NET Core, EF Core, Dapper, SignalR, Hangfire, OpenTelemetry |
| Data | PostgreSQL · Oracle 23ai · Redis · PostGIS · SQL Server: modelling, query tuning, CDC |
| Quality | xUnit v3 · Testcontainers · FsCheck · NBomber · BenchmarkDotNet · Playwright |
| Platform | Docker · GitHub Actions · OCI |
| Certified | OCI Architect Associate · OCI Foundations Associate · Oracle AI Database Foundations |
All repositories use synthetic data only. No employer or customer code, schemas or data appear on this profile.