I build data and AI infrastructure at scale, and I care about how systems actually behave under load: not just that they run, but why they stall, where state lives, and what breaks at the tail. My work sits at the layer between raw event streams and the services that depend on them, covering ingestion, durable orchestration, and turning AI/LLM models into reliable production systems.
- 📊 I build and own data pipelines processing 100M+ events/day, running CDC ingestion with Debezium + Kafka/MSK into Amazon Redshift and holding sub-minute replication lag at that volume.
- 🔧 I debug production at the systems level. I diagnosed and remediated buffer-lock contention and JVM heap exhaustion under sustained load, turning a recurring P0 outage into a non-event.
- 🤖 I build LLM evaluation & serving infrastructure: FastAPI backends, Langfuse tracing for per-run latency and cost, and RBAC-scoped APIs, shipped through CI/CD to ECS.
- 📦 I authored helix-ir (PyPI): schema inference via Algorithm R reservoir sampling, a type lattice with join/meet/subsumes, 1NF normalization, and multi-dialect DDL generation across six warehouses.
- 💻 More on my Portfolio and technical writing on DEV.
Languages: Python · Go · SQL · TypeScript · C++
Data & Streaming: Apache Kafka · Amazon MSK · Debezium (CDC) · Amazon Redshift · Databricks · Apache Spark · Airflow
Backend & Workflows: FastAPI · Gin · REST APIs · Temporal · PostgreSQL · MongoDB · Redis
AI/ML: LLM evaluation infrastructure · Langfuse · ONNX Runtime · embeddings · Computer Vision (YOLO, Faster R-CNN)
Cloud & DevOps: AWS (ECS, Lambda, S3, IAM, EMR) · Docker · Terraform · ArgoCD · CI/CD
- Database & infra internals: how Kafka's log, partitions, and consumer groups actually work; Redis data structures and eviction; Postgres MVCC, indexing, and query planning.
- Consensus & durable execution: implementing Raft from first principles, reasoning about leader election, log replication, and what actually survives a mid-write node failure.
- Lakehouse internals: Apache Iceberg and Delta Lake table formats, and how query engines plan, prune, and commit under the hood.
- Systems performance: profiling latency at p99, backpressure, and where the bottleneck really is versus where it looks like it is.
