# About Dewei Zhai

Dewei Zhai is an independent platform and data engineer based in the Netherlands.
Project-based, implementation-focused, 30-minute first call is free.

- **Location**: Netherlands (PR)
- **Timezone**: Europe/Amsterdam (CET/CEST)
- **Engagement**: project-based freelance — no retainers
- **Domain**: https://zhaidewei.com
- **Email**: dewei.zhai@gmail.com

## Specialties

- Datalake / Lakehouse architecture (Medallion, Unity Catalog, Iceberg, Snowflake)
- Idempotent, config-driven ETL pipelines that survive replays, late data, and timezone edge cases
- 100% Infrastructure-as-Code platforms (Terraform, AWS CDK, CDKTF)
- CDC ingestion (DMS-based and stream-based)
- Cross-cloud data architecture (AWS + Azure + Alibaba Cloud + GCP)
- Real-time and event-driven serverless data pipelines
- AI-augmented engineering practice (custom agents, harness engineering)

## How I engage

A new client gets the first 30 minutes free — long enough for me to understand
the problem and for both sides to decide whether it's a fit. There's no
expectation either way after that call.

Beyond the intro, I bill hourly. I don't do retainers — they tend to drift
into "always-on availability" which I'm not selling. I'm open to **long-term
advisor** engagements where the shape is "regular, bounded check-ins on a
real problem," and I'm fine signing a longer contract under that frame.

There's no minimum project length. I've taken jobs that were a single
afternoon of pairing on a thorny migration, and I've taken jobs that were
twelve months of platform work. The shape of the problem decides, not the
clock.

For exact rate, weekly capacity, and what I have free right now, the live
operating parameters above are the source of truth — I keep them current.

## How I work

I live in the Netherlands, so my default working hours are CET/CEST. Any
**recurring** meeting has to fit that window. I'll take an off-hours call
when a specific project genuinely needs it — but as the exception, never
the routine.

All my current engagements are **remote-first**. I'll consider onsite work,
but more than two onsite days a week as a regular cadence is a non-starter.

On the sync ↔ async axis: the first ~2 weeks of a new engagement I lean
sync — onboarding shortcuts beat misunderstanding budgets. After that I
move strongly async. My default is **about 15 minutes of scheduled meetings
per 8 working hours** — enough for an unblock or a tight checkpoint; not
enough for daily standups, design-by-committee, or vibes management.
Project kickoffs, hard milestones, and wrap-ups don't count against that
budget — those get whatever they need.

## What I'm obsessing over right now

The thing I've spent the most non-paid time on this past year:
**advanced Claude Code usage** as a delivery system.

Concretely: hand-built skills, harness hooks for state-machine
enforcement / notifications / linting, using DeepSeek-powered
subagents (via `claude-ds`) to keep cost down on tool-heavy work,
tightening `CLAUDE.md` files to be densely informational rather than
verbose, and the security posture around what an agent should and
shouldn't be allowed to do automatically.

The ~3× delivery throughput at Enyquant isn't a single trick — it's
the result of doing all of the above and treating "what does the
fastest human + machine pair look like here?" as a real engineering
problem.

New entries in my toolbox over the past year: **Claude Code** (as a
discipline, not a chatbot), **Alibaba Cloud** (forced by the EU-↔-China
split at Enyquant; now genuinely fluent), and **Rust** (slowly,
deliberately, on side projects).

## Related

- Services: /services.md
- Cases: /cases.md
- Stack: /stack.md
- Contact: /contact.md
- Agent index: /agent
- LLM crawler entry point: /llms.txt