Data Warehouse · Big Data · Data Engineering

Jiaxin Zhang

Resume Snapshot

Data Engineering & Data Warehouse Profile

  • Profile Second bachelor's degree in Data Science and Big Data Technology from China Agricultural University; targeting Data Engineer and Data Warehouse Engineer roles
  • Production 10 months of enterprise data warehouse experience at Muyuan's headquarters, working with MaxCompute, DataWorks, Hologres, and SQL
  • Warehouse Dimensional modeling, layered transformations, dependency governance, data-quality validation, and incident tracing
  • Streaming Reproducible hands-on work with Flink SQL, Kafka, Paimon, Doris, and batch-versus-streaming metric reconciliation
  • AI Data Governed Novel synthetic-data pipeline and Mini-C4 web-corpus preparation pipeline
View Projects Contact Me

Selected Projects: Engineering Evidence, Not Just Claims

Featured Project Doris Flink Paimon Metric Reconciliation

E-commerce Batch & Streaming Lakehouse

A reproducible warehouse that computes the same e-commerce metrics through two paths: MySQL + Doris for batch layers, and Kafka + Flink SQL for a real-time DWD → DWS → ADS chain. A Statement Set materializes Paimon layers and Doris ADS outputs in parallel before field-level reconciliation.

2Batch + streaming paths
6Core commerce metrics
1/7/30Fixed reporting windows
PASSField-level reconciliation
  • ProblemBatch reports and live dashboards can diverge when metric definitions, windows, or source paths differ.
  • DesignDoris implements physical batch layers; Flink SQL carries shared definitions through temporary views and writes Paimon plus Doris outputs in one Statement Set.
  • ResultAll core metrics and average order value reconcile across fixed 1-, 7-, and 30-day windows.
Architecture for the e-commerce batch and streaming warehouse
View project details ->
SYNTHETIC DEMO AI Data Engineering Data Governance v0.1.0-demo

Governed Novel Data Pipeline

Lessons from a private three-person AI writing workspace exposed gaps in rights traceability, schema stability, and semantic validation. I rebuilt the pre-training portion as a 100% synthetic, publicly auditable data-governance pipeline.

17/17Automated checks passed
23SHA-256 verified artifacts
20Unique final works
0/0/0Author / work / text-hash overlap
  • ProblemSchema drift and feature collapse passed structural checks that were incorrectly treated as semantic proof.
  • RebuildRights gate, normalization, exact and near-duplicate removal, author/work isolation, dataset production, and version lineage.
  • EvidencePinned commit, dual-version CI, reproducible builds, checksums, Dataset Card, and verification report.
TRAINING-DATA CONTROL PLANE SYNTHETIC DEMO
01Rights
Gate
02Text
Normalization
03Exact / Near
Deduplication
04Author / Work
Isolation
05SFT / Preference
Evaluation
06Public
Release Gate
Manifest & version lineage
Tests & isolation assertions
SHA-256 artifact verification
View audit, rebuild, and evidence ->
LLM Data Mini-C4 Validation
Three iterations of the Mini-C4 data pipeline

Mini-C4 Web Corpus Pipeline

Reproduces and extends the open-source Mini-C4 workflow on a single CPU / WSL machine: WARC parsing, main-text extraction, cleaning, deduplication, language routing, quality filtering, training packaging, and consistency checks.

10Pipeline stages orchestrated
14/14Consistency checks passed
View project details ->
Production Case DataWorks Lineage Review Scheduling Governance

Production Case: Governing a Twice-Daily Dependency Chain

Restructuring a mixed-frequency schedule

For a chain combining self-dependencies, cross-day dependencies, and mixed schedules, I mapped lineage and configuration first, split the change surface by core, transformation, and cache layers, then validated row counts, business metrics, and historical backfills.

01 Classify dependencies Label weekly, daily, and hourly cycles plus normal, self, and cross-day dependencies.
02 Separate risk layers Distinguish core tables, transformation tables, and cache layers to control the blast radius.
03 Isolate time windows Separate early-morning and afternoon runs while aligning cross-day boundaries.
04 Validate three ways Cross-check volume, business metrics, and historical backfill results.

Delivery scope and incident handling

  • ScopeAdjusted upstream tables and dependent chains for twice-daily delivery.
  • PipelineChanged ODS extraction frequency and intermediate transformations, validating each upstream/downstream layer.
  • IncidentLocated and fixed a case-sensitivity mismatch in transferred-group data.

Technical Capability Map

01

Cloud Warehouse

Production cloud warehousing and orchestration

  • MaxCompute
  • DataWorks
  • Hologres / SQL
02

Model & Quality

Layered modeling and data-quality governance

  • ODS / DWD / DWS / DM
  • Dimensional modeling
  • Quality validation / incident tracing
03

Realtime Practice

Reproducible streaming practice

  • Flink SQL / Kafka
  • Paimon / Doris
  • Metric contracts / dual-path reconciliation
04

Engineering

Reproducible engineering delivery

  • Python / Pandas
  • Docker / WSL2
  • Testing / documentation / lineage

Credentials & Public Evidence

Certifications establish a knowledge baseline; merged open-source work and public material add independently verifiable evidence of collaboration and business communication.

Professional Certifications

Professional certifications relevant to target roles

Certifications support the profile; public repositories, run records, and validation results remain the primary engineering evidence.

Jiaxin Zhang's Alibaba Cloud ACP Big Data certification
Alibaba Cloud ACP · Big Data

Alibaba Cloud Certified Professional — Big Data

Knowledge credential supporting hands-on cloud warehouse experience and portfolio evidence.

Valid through Nov 15, 2027 View original certificate ->
Jiaxin Zhang's Alibaba Cloud ACP Large Language Model certification
Alibaba Cloud ACP · LLM

Alibaba Cloud Certified Professional — Large Language Model

Supports the LLM foundation represented by the Mini-C4 and pre-training data-governance projects.

Valid through Jan 16, 2028 View original certificate ->

Technical Writing: From Incident Diagnosis to Durable Practice

Two public write-ups cover real-time pipeline troubleshooting and file-based state management for AI tools, with explicit boundaries, validation methods, and reusable conclusions.

Streaming Data Engineering Flink CDC · DataHub

Alibaba Cloud Flink CDC × DataHub: A Closed-loop Troubleshooting Guide

Reviews six synchronization and consumption failures by separating control plane, data plane, event semantics, and validation signals into a reusable diagnostic sequence.

Read on GitHub ->
AI Engineering Practice OpenClaw · Codex

File-based Memory for OpenClaw and Codex

Stores task state, incident logs, shared knowledge, and permission boundaries in portable files so the same operating method can be reused across OpenClaw and Codex.

Read Chinese-language article ->

Contact

Open to data engineering and data warehouse opportunities where reliable pipelines create measurable business value.

Phone 176-1112-0295 Email 1611209560@qq.com GitHub github.com/JiaxinZhang-space