Job Description

We are looking for a full-stack engineer who has built production applications, and who has also helped build software that other engineers then built on top of — a shared library, an internal tool, a reusable component set or a software development kit — establishing engineering standards through automation rather than documentation.
The role combines application development with internal platform and developer-tooling engineering, and the two feed each other: building a real application end to end is how we discover which capabilities the platform should own.
On the platform side you will build GE HealthCare's internal AI engineering framework — the shared libraries, code generation tools, standardised interfaces and automated verification our data scientists and engineers use to take their own AI solutions into production. It exists to deliver four outcomes for the company:
• reduce time from working prototype to production from months to days;
• allow data scientists to ship production-grade AI without expertise in cloud infrastructure, deployment, application security or front-end development;
• make security, compliance and consistency automatically enforced properties of every solution, rather than outcomes dependent on scarce expert review;
• allow underlying technologies to be replaced — model provider, agent framework, user interface approach — without rewriting the solutions built on them.
On the application side you will develop selected AI applications end to end — front end, back end and agents — for priority business use cases, and for the reference implementations that prove each platform capability before it is offered to others. Expect the balance to sit somewhat more on the platform side than the application side, and to shift as the platform matures. It is not an infrastructure operations role.

tex

GE HealthCare's Chief Data and Analytics Office delivers data, insight and AI products across Finance, Commercial, Supply Chain, Quality, Manufacturing and Operational Excellence.

Most people creating AI solutions here are data scientists and analysts rather than career software engineers. Their modelling, evaluation and domain expertise is what these programmes need; deep infrastructure and front-end expertise is not reasonable to require of them. Every solution must nonetheless reach production as a secure, reliable, supportable enterprise system. A defining characteristic of our approach is that correctness is established by automated tooling — type systems, generated code, schema validation, pipeline checks — rather than by expert human review, which does not scale to the pace required. Your users are colleagues, and your work is measured by how much they accomplish correctly and independently.

Core Responsibilities

Build shared platform capabilities

• Reusable libraries used across AI applications: configuration, structured logging and tracing, error handling, resilience, messaging, data access, API scaffolding, authentication, and access to cloud and AI services.
• The shared execution layer for AI agents — the runtime that takes an agent's declared configuration, instructions and business logic functions and supplies everything else needed in production: lifecycle, messaging, retries, tracing, permission enforcement, output evaluation and deployment.
• The configuration-driven user interface capability, in which a business or data science user describes a dashboard in a structured data file and the platform renders it, so no application code is written per view — including the typed, validated visual components it draws from and the design token system that keeps every view consistent and accessible.
• The data access layer that resolves a declared dataset into a validated result, applying security filtering, parameter validation, resource limits, caching and observability centrally rather than in each solution.

Build applications, where that is the fastest way to prove the platform

• Develop selected AI applications end to end — user interface, API, data access and agents — as reference implementations and early pilot use cases, and harvest from them the capabilities the platform should own. Shared capability here is derived from working code with real consumers, not designed speculatively.

Build the tooling that makes standards real

• Code generation tools and project templates (for example copier or cookiecutter) producing complete, standards-compliant starting points needing no manual correction — including the mechanism by which existing projects adopt later improvements instead of diverging.
• Automated verification in the delivery pipeline — formatting, strict static type checking, architectural boundary rules, interface and schema compatibility, migration checks, security and dependency scanning, automated quality evaluation of AI output — built once as shared components used by every project.
• The machine-readable interfaces between components and roles (for example OpenAPI, JSON Schema, Zod), and the checks that detect when either side has drifted.

Maintain and evolve the platform — a core responsibility; it will be in continuous use and change for years, and its value depends on staying dependable while it changes

• Manage published interfaces: semantic versioning, change logs, migration paths, deprecation, and compatibility tests that catch a breaking change before it reaches a consuming application.
• Assess, validate and complete automated dependency upgrade proposals (for example from Renovate or Dependabot).
• Improve error messages, documentation and reference examples continuously. Where a colleague had to ask a question, treat the question as a defect and remove the need for it.
• Contain scope: add shared capability only where repeated need is demonstrated by more than one real consumer, keeping what a solution author must write as small as possible.

Collaborate and support

• Support colleagues using the platform, converting recurring difficulties into tooling, checks, clearer messages or documentation; review code substantively and document design decisions and rationale.
• Express each service's infrastructure, permission and operational requirements as version-controlled declarations that satisfy automated validation, working with infrastructure specialists on anything not yet available.
• Use AI coding assistants (for example Claude Code, Copilot, Cursor) effectively and responsibly — scoping work, supplying context, remaining accountable for the output.

Experience & Qualifications

• Bachelor's degree in Computer Science, Software Engineering, IT or a related field, or equivalent demonstrable practical experience. An advanced degree is welcome but not required.

Software engineering

• Strong Python, with static type checking (mypy, Pyright) and runtime validation (Pydantic) as a normal part of development rather than an added burden; working TypeScript and a modern component-based front-end framework (React, Vue, Angular) with an understanding of strict type safety.
• Well-structured HTTP API design — specification (OpenAPI), versioning, backward compatibility — and relational databases and SQL: schema design, migrations, query behaviour, parameterised queries as an ingrained habit.
• Version control, code review and CI/CD, including authoring and debugging pipeline configuration (GitHub Actions, GitLab CI, Jenkins); automated testing at unit, integration and interface-contract level, with judgement about which suits a given risk.

AI and Generative AI engineering

• Practical experience integrating large language models into production software: instruction design, tool and function calling, structured output, and management of cost, latency and failure modes. Retrieval-augmented generation in practice — embeddings, vector search (pgvector, FAISS, OpenSearch, Pinecone), retrieval quality.
• Experience with at least one AI agent framework (for example LangGraph, LangChain, AWS Bedrock AgentCore or CrewAI), and the ability to build software that uses it without becoming dependent on its specific types.
• Evaluation of non-deterministic systems. You have defined test cases and quality measures for a system whose output varies — with a framework (Ragas, DeepEval, LangFuse) or a purpose-built harness — and used the results to make a release decision. Automated evaluation is a mandatory gate before deployment here.

Cloud and delivery engineering Infrastructure and production operations are specialist disciplines owned elsewhere in the organisation; our applications request what they need through version-controlled declarations validated automatically, rather than by authoring infrastructure or access policies directly. You are not expected to author infrastructure modules, access policies or network components, nor to own cloud estate design, release execution or infrastructure on-call. Required at this depth:

• Containerisation (Docker), and working knowledge of a major public cloud, ideally AWS, at service level — serverless compute (Lambda), object storage (S3), messaging (SQS), secret management, managed databases (Aurora, RDS), managed AI model services (Bedrock) — sufficient to design against them and reason about behaviour, cost, quota and failure.
• Read and review infrastructure-as-code fluently (Terraform, Pulumi, CloudFormation), and understand access permissions at the level of consequence: what a grant allows, and when a request is broader than the task requires.

Operational literacy — interpreting distributed traces and structured logs (OpenTelemetry, Datadog, Grafana), defining service level objectives and runbooks — and awareness of delivery security practice: immutable artefacts, build provenance, secret scanning, dependency vulnerability management, short-lived federated credentials