Backend Software Engineer | Java · Spring Boot · Distributed Systems · AWS

I build and troubleshoot backend systems where transaction correctness, concurrency, performance, and failure recovery matter. My primary work spans Spring Boot services, event-driven architecture, data-intensive APIs, and AWS-based delivery. I also build across the full stack when the problem calls for it.

LinkedIn · Engineering portfolio · Email · All repositories

Software Engineer at Linux Socials (Apr 2025–present) · Previously, Merkle Inspire (2021–2024).

Selected backend engineering case studies from my portfolio:

• Fraud screening microservice — Java/Spring Boot, idempotent request handling, Redis lookups, bounded rate limiting, circuit breakers, and incident observability.
• Financial transaction workflows — Spring transactions, isolation levels, optimistic locking, SQL/index tuning, cursor pagination, and caching.
• Job-feed integration portal — React + Spring Boot, parallel external API calls, bounded retries, batch persistence, and cache-backed refreshes.

Selected outcomes documented in my portfolio: 40% lower p95 latency from profiling and persistence-path tuning, and mean time to recovery reduced from 35 to 12 minutes through operational improvements. These describe specific production work, not general benchmarks.

• NUMEN — Java-backed data-intelligence workflows that collect from explicitly permitted public sources, preserve provenance, validate records, and run via Docker Compose. Focus: auditable data flow, failure visibility, and one-step startup.
• Aegis Insurance Decisioning — Simulated insurance operations with a Java/Spring Boot backend, Angular frontend, risk-rule decisions, and audit trails. The documented demo uses an in-memory data store.
• Vortex CUDA — Exact GPU vector search spanning CUDA, C++20, Java JNI, and Spring Boot. On the documented RTX 3050 workload, the project reports 48/48 benchmark validations, 50% less FP16 vector storage, and ~16.7× lower p50 latency vs its own scalar CPU reference—not a production vector database. Methodology · Hardware and repeated-run evidence.

More engineering experiments: BedRock (early Go orchestration prototype with verification and rollback) · CHAIN//REACTION (synthetic causal-resilience simulation) · Distributed transaction lab (fork; MVCC and clock/conflict experiments).

5 merged engineering PRs across 4 external upstream projects — Apache Fluss, Apache Pulsar, Apache Druid, and AutoMQ. I focus on correctness, concurrency, performance-sensitive dispatch, and failure recovery. Only merged code contributions authored by BackendArchitectX are included here; repository-local and domain-registration PRs are excluded.

• Apache Fluss · RPC endpoint correctness (#4263) — Problem: caching connections by server UID alone could reuse an outdated endpoint after a host/port change. Change: keyed connections by UID, host, and port; made readiness and disconnect operations endpoint-aware. Added regression tests for rolling-upgrade and connection-lifecycle behavior.
• Apache Pulsar · Namespace-indexed topic watchers (#26969) — Problem: topic notifications scanned listeners registered for unrelated namespaces. Change: indexed persistent-topic watchers by namespace and dispatched callbacks outside the registration lock. Added concurrency and listener-lifecycle tests plus a JMH comparison with the original linear scan.
• Apache Druid · Preserve user error classifications (#20633) — Problem: errors from data servers could lose their INVALID_INPUT / user-error classification. Change: preserved exception causes in multithreaded groupBy execution and reconstructed structured DruidException responses while retaining legacy compatibility. Added regression tests across both paths.
• AutoMQ · Remove redundant topic cleanup (#3542) — Problem: duplicate deleted-topic traversal produced redundant coordinator cleanup notifications. Change: removed the extra traversal, retaining callbacks for actual partition deletion and state transitions.
• AutoMQ · Controller-only metrics startup (#3493) — Problem: metrics-reporter initialization could run on controller-only processes. Change: skipped broker-only configuration, producers, and reporter threads when the broker role is absent; covered broker, controller, and combined-role cases.

Distributed systems: Trino #30973 (transaction-commit race) · Apache Pinot #19915 (replica-group lookup hashing) · Apache Fluss #4230 (custom Paimon lake-table paths).

Reliability and security: AutoMQ #3579 (Prometheus Basic authentication) · AutoMQ #3555 (reader-close lifecycle) · Nasiko #343 (Windows queue writes).

PR statuses verified November 2, 2026. Open proposals are not counted as merged contributions.

Core strengths: Java 8/11/17 · Spring Boot · Microservices · Distributed Systems · AWS · Kafka · SQL · Redis

• Java & backend: Java/J2EE, Collections, Streams, multithreading, Spring MVC, Spring Data JPA, Hibernate/JPA, Spring Security, JSP/Servlets, REST APIs, DTOs, validation, OpenAPI/Swagger, Postman.
• Architecture & messaging: Kafka, IBM MQ, RabbitMQ, asynchronous processing, transactional boundaries, idempotency, rate limiting, retries/backoff, circuit breakers, caching, SOLID, design patterns.
• AWS & DevOps: EC2, EKS, RDS, S3, CloudWatch, Docker, Kubernetes, PCF, Jenkins, GitHub Actions, Maven, Gradle, Git, GitHub/GitLab/Bitbucket.
• Data & observability: MySQL, PostgreSQL, Oracle SQL, Redis, SQL tuning, indexing, pagination, optimistic locking, JFR/JVM profiling, Splunk, Dynatrace, Grafana, logging, metrics, incident investigation.
• Full-stack: React, TypeScript, JavaScript, HTML5, CSS3, JSP, Angular (project experience), and Java-to-frontend API integration.
• Security & quality: OAuth2/JWT, authentication/authorization, JUnit 5, Mockito, TDD, unit/integration testing, regression tests, negative-path tests, CI/CD and code reviews.
• Additional tools & languages: Python, C++; FastAPI/Flask, Node.js/Express and MongoDB (working exposure); Go (learning through prototypes); GitHub Copilot, Cursor, Claude and AI-assisted workflows with human review.

Build for failure. Measure before optimizing. Preserve the invariant.

I favor explicit ownership, simple designs, reproducible tests, and performance claims supported by workload-specific evidence.

Engineering investigation notes — CI test discovery and search semantics, documented separately from merged upstream contributions.

Open to backend engineering opportunities and thoughtful open-source collaboration.

GitHub contribution activity · Portfolio · LinkedIn · Contact