I build data platforms, and the AI agents that run on top of them.
I care about systems that are easy to operate, easy to explain, and boring in the places where reliability matters. Most of my work sits where data products, AI tooling, and engineering platforms meet — and most of it ends up open source.
These days most of what ships here is written alongside coding agents, with me steering. I describe intent, review diffs, and keep the architecture honest — the agents do the typing, the searching, and a lot of the grunt work.
My default pair — reads the whole tree, reasons about design, and writes the load-bearing code.
Hand it a goal and a repo; it plans, edits across files, and runs the tests until green.
Open-source TUI agent wired to homelab models for quick, private edits without the cloud.
Data pipelines, warehouses, and observability. Agent workflows, model routing, evaluation, and usage analytics. Small, sharp tools with clean interfaces.
Python, Rust, TypeScript · Spark, Airflow, ClickHouse, BigQuery, Kafka · Kubernetes, AWS, GCP, Cloudflare · LlamaIndex, AI SDK, LangGraph.
Areas of expertise
8+ years of delivery across data, AI, and platform engineering — here's where the time actually went.
Data Engineering
8 yrEnd-to-end pipelines from raw event streams to analytics-ready tables. ClickHouse as the primary OLAP layer — petabyte-scale ingestion, columnar query optimization, and materialized view patterns. Airflow for orchestration; Spark for heavy batch transforms.
AI Agent Engineering
4 yrAgent frameworks, model routing, and evaluation infrastructure. AnyRouter for multi-provider fallback and BYOK routing. LangGraph and LlamaIndex for workflow orchestration. Usage analytics baked in from day one.
Cloud Infrastructure
6 yrKubernetes clusters on AWS and GCP, Helm chart authoring, Terraform for IaC. Cloudflare Workers for edge compute — zero cold-start, global, cheap.
Backend & APIs
7 yrREST and GraphQL services in Python, Rust, and TypeScript. Cloudflare D1 + KV for edge-native persistence. Durable Objects for stateful agent sessions.
Frontend & UI
5 yrReact and TanStack Start with SSG. shadcn/ui + Tailwind as the component layer across 8 sibling apps. Performance-first: sub-second LCP, no hydration overhead for static routes.
DevOps & Observability
5 yrGitHub Actions CI/CD with Turborepo caching. Structured logging into ClickHouse; MotherDuck for ad-hoc querying. OpenTelemetry traces for agent workflows.
Open Source
8 yrPublic libraries, dashboards, and reference implementations. rust-tieng-viet, ClickHouse Monitor, ShareHTML, LLM Timeline — built and maintained openly.