Computer Science & Data Science @ Rutgers University
Pragmatic Backend Engineer focused on Distributed Systems, APIs, and Data Pipelines
Backend and Data Engineer with a strong foundation in building high-throughput data pipelines, distributed systems, and scalable backend infrastructure.
Proficient in Python and Go, using frameworks like FastAPI and Gin, with strong SQL experience across PostgreSQL, MySQL, and MongoDB, plus streaming and orchestration experience with Kafka, Redpanda, and Airflow. Comfortable working across AWS and GCP for deployment and infrastructure.
My work centers on pipeline performance optimization, event-driven architecture, and building systems that stay reliable under real load.
Python Go SQL Java TypeScript JavaScript Bash
FastAPI Gin LangChain Airflow Django Node.js Express gRPC Celery
PostgreSQL MySQL pgvector Redis MongoDB
AWS GCP Docker Kubernetes Kafka Nginx GitHub Actions Linux Prometheus Grafana
Go • Gin • PostgreSQL • Kafka • Redis • Docker
A horizontally scalable, durable job queue in Go where Postgres is the single source of truth. A job is safe the moment its insert commits, so crashed workers, broker restarts, and dropped wake-ups cost latency but never work. Kafka and Redis sit on top as optional fast paths.
Go • gRPC • Python • LightGBM • Prometheus • Docker
A distributed cache that uses a machine learning model to decide what to evict, approximating the optimal algorithm instead of falling back on LRU. The model runs inside the eviction path with a sub-microsecond budget, which forced me to use flattened tree layouts, sharded locking, and zero-allocation request handling.
Python • Redpanda • Airflow • PostgreSQL • Pydantic • Docker Compose
An e-commerce event pipeline where a producer publishes clickstream events to Redpanda, a consumer validates each message and batch-loads it into Postgres, and Airflow runs the whole thing as a DAG that ends in funnel and item-popularity summary tables. Invalid messages go to a dead-letter table with the error attached instead of disappearing.
I built it to learn the orchestration and streaming split firsthand, and it taught me two things I now design for up front: Airflow tasks have to terminate, so the consumer drains a bounded batch instead of polling forever, and re-runs duplicate data unless the transform is idempotent, so it rebuilds from deduplicated rows in a single transaction.
Email: justinkong.dev@gmail.com Portfolio: justinkong.app LinkedIn: linkedin.com/in/justin-hkong