Tìm việc làm IT

VinFast is a pioneering electric vehicle (EV) company committed to revolutionizing the automotive industry with sustainable and innovative mobility solutions. Within the Automotive AI organization, the AI Personalization team builds the layer that makes a virtual assistant feel like it knows the person it is talking to: a long-term user memory service and a proactive agent that decides what to say, when to say it, and — just as often — when to stay silent. It sits on a full conversational stack: LLM-based reasoning and tool use, speech in and out, audio front-end processing in noisy real conditions, and an interface spanning voice and screen. The team ships capabilities other product teams consume, not a screen of its own. Your users are therefore both end users and other engineering teams, and your definition of done is a contract someone else builds on top of. The Senior Product Owner / Product Manager owns the backlog, acceptance criteria and delivery cadence for a feature group in this layer: turning a vague ambition — “say the right thing at the right moment, and know when not to speak” — into testable requirements with numbers attached, negotiating the data and service contracts that make them possible, and driving delivery through hard acceptance gates. Scope grows with the person: a product owner who can both hold a technical position and run a clean delivery will be given a new problem area to own outright. Our assistant runs in cars. We do not expect you to arrive knowing anything about electric vehicles — that will be coached, and it is not what we are hiring for.

• Own the capability as a product. Backlog, roadmap and the contracts offered to consuming teams. Decide what gets built, declined and deferred — and defend it in writing.
• Own assistant behavior. When the assistant acts automatically versus suggests versus asks; how confidence is gated and when it should abstain; how it opens a conversation the user did not start; what it does on cold start, when uncertain, and when wrong. Work with UX/CX on the interaction contract across voice and screen — turn-taking, interruption, error recovery, and how much the assistant may reveal that it remembers.
• Own the quality bar. Grounding, tool-calling boundaries, guardrails, and the evaluation set — what “good” means, who labels it, how regression is caught before release. Set and defend thresholds on the components you depend on: recognition accuracy under real noise, latency, intent accuracy, recommendation precision. Know why a strong offline metric can still be a bad outcome.
• Turn requirements into acceptance. Convert business objectives and use cases into system requirements and acceptance criteria with explicit numeric thresholds — accuracy, latency, coverage, false-trigger and abstain rate, cold-start behavior. Own the Definition of Done and maintain traceability from stakeholder requirement through architecture to system test, in line with the program’s process framework (Automotive SPICE, SYS.1–SYS.5), including readiness for formal assessment.
• Run delivery. Sprint planning, release plan, dependency register, critical path and risk log, defect triage, QC entry and exit criteria, hand-over at acceptance milestones. Raise slippage early enough that it is still a decision rather than an announcement.
• Close cross-team contracts. Data contracts with the platform team (schema, freshness, frequency, volume budget), service contracts with the cloud team, deployment contracts with the on-device team, acceptance contracts with product experience and validation. Convert “we’ll look into it” into an owner, a scope and a date.
• Own the product side of privacy. Personalization runs on personal behavioral data & at times on voice: what & why is collected, retention, consent, deletion, and the boundary of what may be inferred or surfaced Requirements Education & experience
• Bachelor’s degree or higher in Computer Science, Software Engineering, Data Science, Electronics/Telecommunications, HCI or a related discipline.
• 5–8 years total professional experience, with at least 3 years as a Product Owner / Product Manager / Technical Program Manager on a product that shipped to production. Earlier time as an engineer, data scientist, UX designer or QA counts toward the total.
• Mandatory — depth in at least one of: conversational AI / virtual assistant products; LLM or agent systems (RAG, tool use, evaluation, guardrails); speech and audio (ASR, TTS, wake word, audio front-end — noise, echo, speaker separation); personalization or recommendation systems; platform or API products consumed by other engineering teams. Depth in one is what we look for; breadth is a bonus, not an expectation.
• Automotive, EV, embedded and IoT experience is a plus. Technical foundation — screening gate, pass/fail You will spend your week in rooms full of architects and ML engineers, without a translator. You must be able to:
• Read, critique and write a service contract, a data schema and a sequence diagram.
• Explain the lifecycle of the AI system you have worked on, and where quality is lost along it.
• Reason quantitatively about an end-to-end latency budget, and about what belongs on-device versus in the cloud.
• Pull and sanity-check your own numbers — working SQL, comfort with a BI or notebook tool.
• Hold a position in an architecture review: disagree with a technical reason, and change your mind when given a better one. Coding is not required. Being unable to follow the discussion is disqualifying. Product ownership & delivery — primary ranking criteria
• Ownership of a backlog from problem statement to release, with acceptance criteria a QA engineer can execute without asking you a question.
• Advanced Jira and Confluence; sprint mechanics, dependency and risk management, release planning, defect triage.
• A track record of negotiating scope — you have cut or deferred a committed feature, secured it in writing, and made it hold.
• Experience driving external teams, who do not report to you, to dated commitments.
• Certifications (CSM/PSPO, PMI-ACP, PMP, SAFe) are welcome but do not substitute for this evidence. Soft skills & languages
• Writes short, precise, unambiguous documents. A one-page decision note beats a forty-slide deck.
• Influences without authority; stays factual under pressure — names gaps, not individuals.
• Holds a position against senior stakeholders when the data supports it, and concedes cleanly when it does not.
• Vietnamese native or fluent; English professional working proficiency — reading technical specifications, writing status and design notes, presenting to international partners.