Forward Deployed Engineer, GenAI (Thai, English)

What matters most

• Ship agentic solutions into production

  The role calls for coding and debugging customer-specific workflows, including multi-agent systems and MCP servers, from prototype through launch, with measurable ROI.

• Integrate AI with customer environments

  Architect and code connections between Google AI products and customer APIs, legacy data silos, and security perimeters. The qualifications also emphasize structured and unstructured data pipelines, vector databases, and RAG-like architectures.

• Make performance and safety measurable

  Build evaluation pipelines and observability frameworks for accuracy, safety, and latency. The posting also values knowledge of tokens per second, cost per request, state management, and granular tracing.

• Co-build and communicate across audiences

  Work alongside customer engineering teams, share development practices, and explain technical concepts to non-technical executives and Thai-speaking audiences. Fluency in Thai and English is a stated requirement.

• Turn deployment lessons into reusable improvements

  Identify recurring field friction and convert it into reusable modules or formal product feature requests for engineering teams.

Likely challenges to think through

• Prototype speed versus production reliability

  Preparation hypothesis: rapid agentic prototypes may need substantial work before they meet accuracy, safety, latency, and ROI goals. Prepare to explain how you would define acceptance measures, evaluate failures, and decide when a workflow is ready to launch.

• Customer-specific integration versus reuse

  Preparation hypothesis: legacy data, APIs, and security boundaries may demand bespoke code, while the role also calls for reusable modules. Prepare an approach for separating customer-specific adapters from repeatable components without assuming access to data or infrastructure.

• Agent capability versus operational cost and visibility

  Preparation hypothesis: multi-agent behavior and state can make a system harder to trace and more expensive or slower to run. Prepare to discuss how granular tracing and LLM-native metrics would inform design trade-offs.

• Technical depth versus shared understanding

  Preparation hypothesis: co-building with engineers while explaining outcomes to executives in Thai and English may require different levels of detail. Prepare a concise explanation of one technical trade-off in both an engineering-focused and a non-technical form.

Topics to prepare

• Cloud AI architecture

  The minimum qualifications call for architecting AI systems on cloud platforms such as GCP; the posting names Gemini and Vertex AI. Revision exercise: sketch how a customer-specific application would use a model, data pipeline and customer infrastructure, and explain the boundaries between them.

• Enterprise data readiness and retrieval

  The minimum qualifications specify structured and unstructured pipelines, vector databases and RAG-like architectures. Revision exercise: diagram ingestion, retrieval and answer generation for a hypothetical customer data silo; identify where data quality or access could block deployment.

• Agentic workflow design

  The responsibilities name production-grade agentic workflows, multi-agent systems and MCP servers. ReAct, self-reflection, hierarchical delegation and named frameworks are preferred experience. Revision exercise: compare a single-agent design with a multi-agent design and justify when the added coordination is worthwhile.

• Production integration and state

  The role addresses APIs, legacy data silos, security perimeters and state-management challenges in customer environments. Revision exercise: trace a hypothetical request across those boundaries and identify likely integration, access and state-recovery failure points.

• Evaluation, observability and value

  The responsibilities specify accuracy, safety, latency and measurable ROI; tokens per second, cost per request and granular tracing appear in the preferred qualifications. Revision exercise: propose a small set of quality and operational measures, then explain what each would reveal before launch.

• Customer-facing technical discovery

  The role involves co-building with customer engineers and explaining technical concepts to non-technical executives and Thai-speaking audiences; leading discovery sessions is preferred. Revision exercise: prepare questions that uncover a use case, data constraints and a measurable outcome, then outline separate engineering and executive explanations.

Skills to revise

• Implement and debug in Python

  The minimum qualifications specify software development using Python or a related language, and the role involves coding and debugging. Practice exercise: write a small application that calls an AI service through an API, handles failures and records enough information to diagnose an unsuccessful request.

• Build a retrieval pipeline

  Structured and unstructured pipelines using vector databases and RAG-like architectures are explicit minimum-qualification experience. Practice exercise: ingest a small mixed-format dataset, retrieve relevant passages for test questions and inspect cases where retrieval produces weak evidence.

• Integrate across customer boundaries

  The responsibilities include coding connections between Google AI products and customer APIs, legacy data silos and security perimeters. Practice exercise: build a mock integration with an access check and an unavailable-data case, then document how each failure is surfaced.

• Prototype an agentic workflow

  The role calls for agentic workflows, including multi-agent systems and MCP servers; specific agent frameworks are preferred, not mandatory. Practice exercise: implement a small workflow with a defined handoff or MCP server interaction, and test how it preserves state when a step fails.

• Evaluate and trace behavior

  Building evaluation pipelines and observability frameworks for accuracy, safety and latency is an explicit responsibility. Practice exercise: create a small test set, log per-step outcomes and latency, and use one failed case to explain a proposed change. Include token or request-cost measures as preparation for the preferred metrics knowledge.

• Explain a solution in Thai and English

  Fluent Thai and English communication is a minimum qualification, and the role requires explanations for non-technical executives. Practice exercise: give a brief explanation of the same proposed AI workflow in both languages, stating its expected benefit, a limitation and how success would be measured.

• Prepare a production launch walkthrough

  The role calls for customer AI solutions taken from conception to launch. Prepare one project narrative showing your coding contribution, deployment blockers, collaboration with customer engineers, and measured outcome; distinguish your work from the team’s.

• Sketch a customer-environment architecture

  As a practice exercise, draw a Google Cloud AI solution that connects to customer APIs and legacy data behind security perimeters. Show how structured and unstructured data enter a vector-backed RAG pipeline, and mark data-readiness and integration risks.

• Work through an agentic prototype-to-production design

  The role includes production-grade agentic workflows, with multi-agent systems and MCP servers given as examples. Prepare a small design or code walkthrough explaining tool access, state management, failure handling, and what would change before launch.

• Build an evaluation and observability plan

  Prepare a sample test set, acceptance criteria, and tracing plan for accuracy, safety, and latency. Include how you would inspect failures and track LLM-native measures such as tokens per second and cost per request, which the posting lists as preferred knowledge.

• Practice a customer discovery and feedback scenario

  Rehearse questions that uncover the customer’s use case, data constraints, security boundaries, and definition of ROI. Prepare an example of how a recurring deployment problem could become a reusable module or a specific product feature request.

• Explain the design in Thai and English

  Fluent Thai and English communication is required. Practice a concise technical explanation for customer engineers and a non-technical executive summary in both languages, covering the proposed value, risks, and measures of success.

• Bring targeted questions for the employer

  Ask how the team divides coding and deployment work with customer engineers, measures production success, and routes field insights to product engineering. These are questions to clarify, not assumptions about Google’s process.