To apply for this job, you need to complete both steps below:

STEP 1:

Please click the link to submit your application directly to the company: 

Careers

Your application will only be received by Recruiter if submitted via above link. 

STEP 2:

Kindly scroll to the bottom of this page and complete the short VinUni Tracking Form.

Filling out this form alone does not count as applying. Kindly remind this form is not part of the company’s application process. It only helps Careers, Alumni, Industry and Development (CAID) Department discover more opportunities and follow up in case of system issues.

What you'll be responsible for
1. Data acquisition & integration

• Build ingestion pipelines for: vehicle telemetry, DMS / work-order history, CRM / O2O, charging, service, navigation.
• Work directly with source owners to settle the data contract: grain, keys, cadence, and how a changed record is identified.
• Read from S3, BigQuery and Redshift in batch — hive-partitioned and safe to re-run.

2. Data platform

• Airflow 3 (MWAA) — asset-driven orchestration, no sensors.
• Spark / EMR Serverless — heavy preprocessing.
• dbt (ClickHouse adapter) — staging and mart layers.
• ClickHouse (MergeTree) and an S3 medallion lake.
• Guarantee idempotency, correct partitioning, safe re-runs and backfills, and predictable cost.

3. Data products

• Built in priority order — not all at once:
• Vehicle360 — in progress
• Mobility360 / trips — in progress
• Customer360 — blocked pending the identity spine and consent decisions
• Charging360, Service360 — backlog
• Every data product needs a clear schema, tests, documentation, and an owner.

4. Serving & data access

• For people: dashboards and self-serve views over the marts, so stakeholders stop asking us to run queries for them.
• For AI agents: a governed data-access layer — an MCP server or API over the warehouse — that lets ViTA and other agents fetch context safely, with the right filters, limits and permissions baked in rather than left to the caller.
• Make the semantics legible: an agent querying a mart should not have to guess what a column means.

5. Data quality & governance

• Freshness, volume and schema-drift checks.
• Lineage and metadata at a level that's actually useful.
• PII compliance: biometric sources supply event labels only — never raw images, voice, or embeddings. Cross-brand data requires a lawful basis before it is ever joined.

6. DevOps

• GitLab CI, Docker, AWS (IAM, S3 lifecycle, EMR).
• Logging, monitoring, and alerting when a pipeline breaks
  
  
  
  

Requirements

Must-have

• 3+ years in data engineering, with pipelines you've actually run in production — not personal projects or coursework.
• Strong Python and SQL.
• Apache Airflow — you've written and operated your own DAGs, and you understand backfill and catchup.
• Spark / PySpark — you've handled datasets in the hundreds of gigabytes.
• Data modeling — grain, keys, SCDs, and the difference between a fact and a dimension.
• You can explain: idempotency, partition strategy, late-arriving data, and what you do when an upstream schema changes without warning.
• Practical AWS: S3, IAM, and the ability to read a permissions error and fix it.
• Communication. You'll sit with data owners outside this team. Asking the right question matters as much as writing the code.

Nice to have

• ClickHouse or another columnar OLAP engine: Druid, Doris, BigQuery, Redshift.
• Data visualization — Metabase, Superset, Grafana or similar. Enough to build a dashboard people trust, and enough taste to know when a number needs context rather than a bigger font.
• Building MCP servers or tools for AI agents — exposing warehouse data to an LLM agent as typed, permissioned tools. New for most candidates; genuine curiosity here counts as much as experience.
• EMR / Spark on serverless infrastructure, including job cost tuning.
• Kafka or CDC — not used at this stage, but on the roadmap.
• Vehicle, IoT or telemetry data experience.
• Enough English to read technical documentation.

What we are not asking for

• Stated plainly so you don't rule yourself out: you do not need Scala, Go, Iceberg, Delta Lake, MinIO, Neo4j, Qdrant, MLflow, LangGraph, Kafka
• Connect, MQTT or Debezium. None of them are in the stack today. If we need them later, we'll learn them together.

Who fits here

• Comfortable with ambiguity. You'll hit "the data didn't arrive" far more often than "this query is slow".
• Goes and asks instead of sitting and guessing.
• Treats documentation as part of the work, not an afterthought.
• Would rather leave behind a system that runs than code that's clever.
• Three things changed from the last version: VinFast is now the employer line rather than ViTAI alone; Serving & data access is a full responsibility section (dashboards for people, an MCP/API layer for agents) with a matching bullet in the first-12-months list; and the nice-to-have list gained data visualization and building MCP servers/tools. The agent-facing bullet is deliberately worded to welcome candidates who are curious but haven't shipped one — almost nobody in the VN market has yet, and screening on it as experience would empty your pipeline.