I build correctness-first backend systems — and I'm proving it by shipping a fintech-grade payments & loan-origination platform in Go: idempotent, event-driven, and tested for the failure modes a real payments team is paid to prevent.
Backend-focused engineer (.NET, Go, TypeScript) with a strong CS & mathematics foundation (METU). I build practical, runnable systems and care about the parts that are easy to get wrong — idempotency, concurrency, outbox-style workflows, and failure recovery — backed by real tests, CI, and local verification. My graduation project is a physics-informed neural radiance field in PyTorch. Right now I'm going deep on the slice fintech teams hire for: money-safe ledgers, idempotent APIs, event streaming, and risk analytics.
Event-driven double-entry ledger in Go: idempotent, concurrency-safe transfers (no overdraft under load), gRPC, a transactional outbox + AML consumer, a loan-origination state machine, and a dbt risk-analytics layer. Tested against real Postgres; race-checked, ~88% coverage.
A limit-order-book matching engine in modern C++20: price-time priority, LIMIT/MARKET/IOC/FOK orders, O(1) cancels — ~8.6M orders/sec single-threaded. Property/fuzz-tested (book invariants + quantity conservation), deterministic replay, ASan/UBSan CI.
A Monte Carlo path tracer from scratch in pure Python + NumPy — no OpenGL. Reflection, Fresnel-glass refraction (with total internal reflection), soft shadows, and depth of field, vectorized one bounce at a time.
A vacancy-to-OSS proof engine: maps backend job postings to concrete open-source contribution opportunities, scores fit, and tracks proof actions with idempotency keys — shipped with a UI, CI, Docker, and tests.
A mixed .NET + Go system: an order API emits durable outbox events that a Go worker processes into ledger entries, with end-to-end local verification — the outbox pattern across two languages.
My graduation project: a physics-informed neural radiance field in PyTorch, adding normal-consistency and Lambertian regularizers to recover cleaner geometry from images.
A Go URL shortener service: short links, redirects, click analytics, rate limiting, metrics, and HTTP tests — a compact, production-shaped API.
A Greenhouse-ATS → HRIS/Airtable onboarding integration service: HMAC-verified webhooks, an idempotent task saga, retry/backoff with dead-letter, and a nightly drift reconciler with an append-only audit trail. Solves a real day-one ops problem.
A one-button ink-physics arcade game — fling a gooey ink blob and splatter through the targets. Canvas 2D soft-body rendering, fixed-step physics, slow-mo juice, and procedural Web Audio.
Dependency-free Go validators for Mexican fintech: CLABE interbank accounts (check-digit validate/generate) and payment-card tooling (Luhn, brand, masking, expiry). Table-driven tests, CLI.
A tiny Go money type: integer minor units, exact decimal parsing, and remainder-safe allocation that never loses a cent — proven by a 5,000-case conservation property test.
For fun: an optimized mobile destruction-physics pipeline (Teardown-style voxel destruction) tuned hard for performance and memory on constrained hardware, in a published title.
Failing edge-case tests committed RED, then GREEN — money rounding, idempotency on retry, concurrent double-spend.
GitHub Actions runs lint, vet/typecheck and the full test suite on every push. Badges reflect reality.
Each repo has a DESIGN.md: why this isolation level, why int64 minor units, why idempotency keys.
clone → one command → running. Docker Compose or a single test command, no environment debugging.