Taste is a product skill.
Model output is infinite. Selection is the job.
Portfolio 2026 — Shenzhen / Shanghai
MaxZhang.
AI Product ManagerShenzhen — Shanghai
Building intelligent products with a designer's eye.
Ten years turning model capability into products people trust. 0→1 launches, +80% retention.
- Experience
- 10+yrs
- Shipped
- 0→1
- Retention
- +80%
Scroll to index
Open for select projects
01 / Index
What I bring to the table.
0110+yrsYears shipping productsFrom 0→1 launches to platform scale.
020→1Products launchedAmbiguous model capability → a product people use.
03+80%Retention liftBuilt the loop, not just the launch.
04AI/ MLNative domainEvaluation, guardrails, model-in-the-loop workflows.
02 / Case 01
AI videoproductionpipeline
A media team needed to ship far more video without adding headcount.
Rebuilt the workflow around a model-in-the-loop pipeline: brief → script → generation → review.
Role — Product lead / AI workflow design
+80%Cross-team delivery velocity
¥2.5–3 → ¥0.5–1Cost per video
03 / Principles
A point of view, not a process.
Ship trust, not demos.
A demo shows capability; a product earns the second use.
The model is never the product.
The product is the workflow, the guardrails, the feedback loop.
04 / Method
How I work.
- 01Frame the real jobInterrogate the workflow before touching the model.
- 02Prototype the trustPut the smallest real loop in front of real people.
- 03Instrument the loopMeasure where judgment is lost, not just clicks.
- 04Ship and learnSmall releases, brutal reviews, compounding quality.
05 / Writing
Thinking in public.
All writing01OpenTelemetry Part 1: Why Observability MattersObservability answers the question monitoring cannot: why the system ended up in its current state. This post explains what observability is, how it differs from monitoring, and where OpenTelemetry came from and why it became the de facto standard — for full-stack developers who want a systematic understanding of the concepts and ecosystem.2026-08-0902OpenTelemetry Part 2: The Three Signals and Their Inner WorkingsHow traces, metrics, and logs are each produced and how they relate to each other is the core of understanding OpenTelemetry. This post covers spans and context propagation, metric aggregation and temporality, log correlation, and how the SDK and Collector divide the data path.2026-08-0903OpenTelemetry Part 3: The Ecosystem, Selection, and RolloutAn observability solution consists of three layers: production, collection, and analysis. This post covers OpenTelemetry2026-08-09