About the role
The Forward Deployed Engineer builds working proofs of concept (POCs) in weeks, so that the decision to invest in a production build rests on evidence. You sit with business users in R&D China, understand the problem first-hand, and build something they can try.
Our landscape has several vendor systems, different IT stacks and many small new projects. Vendors build production workloads following AZ guidelines. Your job comes before that: prove quickly whether an idea works, what it takes, and what the risks are, then hand over cleanly to the vendor.
The scope is AI, data and software alike. One month you may connect two vendor systems, the next build a data pipeline or an LLM application. You are supported by enterprise architecture for standards, the cloud team for infrastructure, PMs for process and product owners for direction.
Key responsibilities
Build proofs of concept
Build working POCs for AI, data and software ideas within an agreed time box, typically 2 to 6 weeks.
Agree written success criteria with the product owner before building, and measure the POC against them.
Choose the simplest approach that answers the question: configure, integrate, script or build.
Demonstrate results to users and leaders, and give a clear recommendation: proceed, stop or iterate once more.
Work directly with users
Spend time with scientists, clinical, regulatory and lab staff to see how the work is really done.
Turn loosely defined needs into something concrete that users can react to.
Collect user feedback during the POC and adjust quickly.
Integrate across systems and stacks
Connect vendor systems through APIs, data pipelines and automation.
Explore and profile data early, so that data quality and access problems surface before production.
Work across AWS and Alibaba Cloud, and across the different stacks used by our vendors.
Apply AI where it adds value
Build LLM applications, retrieval and agent workflows, and classic machine learning prototypes.
Evaluate model quality, cost and failure modes, and report them honestly.
Use only AI models and services approved for use in China.
Hand over to production vendors
Know the AZ guidelines for production workloads well enough that POC choices do not block the production build.
Deliver a handover package for each approved POC: code, design notes, data findings, test results, known risks and open questions.
Brief the vendor, support the first weeks of the production build, then step back.
Give the business partner and PM a technical view on vendor scope, design and estimates, based on POC evidence.
Work within guardrails
Build in the sandbox provided by the cloud team, using patterns approved by enterprise architecture.
Use de-identified or synthetic data by default, and obtain approval before using personal, clinical or genetic data.
Keep POCs separate from GxP-validated production systems, and never present a POC as a production service.
Keep all code in the company repository with basic documentation, and retire POCs that are stopped.
How the role works with others
The FDE builds; the surrounding roles set direction, provide the platform and take the result into production.
Role
What they provide
What the FDE gives back
Product owner
The problem, the users, priorities and success criteria
A working POC and a proceed, stop or iterate recommendation
Technology Business Partner
Business context, stakeholder access and the go or no-go decision
Technical evidence for roadmap and investment decisions
PM
Intake, time box, decision gate and handover process
Honest progress, risks and effort estimates
Enterprise architecture
Approved patterns, stacks and AZ guidelines
Feedback on where the patterns help or slow POC work
Cloud team
Sandbox environments on AWS and Alibaba Cloud
Clear infrastructure needs and findings on cost and performance
Vendors
Production build and support under AZ guidelines
Handover package, briefing and early build support
Required qualifications
Experience and…