AI Developer

The EverydayAI Accelerator exists to turn generative AI into working, daily reality across HHMI’s administrative and operational functions. AI Developers design and build the AI systems that actually ship, embedded inside delivery teams alongside engineers, architects, and business partners. 

The foundation here is enterprise product engineering. This role requires real production experience: systems versioned, tested, deployed through CI/CD, and operated under SLAs, with genuine machine learning and deep learning depth built on top. The work is hands-on and end-to-end, from co-designing and building AI systems to deploying, instrumenting, and operating them in production. When something breaks, this is the person who diagnoses and fixes it. 

Why this role matters 

The EverydayAI Accelerator exists to change how HHMI operates, and that change only happens when AI systems actually ship. HHMI has no shortage of ideas. What it needs are engineers who can turn them into working systems. This role sits at the center of that work: building the production systems every Accelerator project depends on, taking a real business problem, selecting the right approach, writing the code, and operating what gets built. The work is technical, consequential, and visible across the organization. 

What you will actually do 

• Build production AI systems with the team. Work inside a delivery team alongside engineers, architects, and business partners to ship production AI. The contribution is collaborative and integrated, not delivered in parallel.

• Pick the right algorithm for the problem. Treat model selection as a design choice. Sometimes the right answer is a large language model with retrieval; sometimes it is a gradient-boosted tree; sometimes it is a well-featured logistic regression with a clean evaluation. Bring real ML and DL fluency to that choice and defend it. 

• Build to the team’s patterns. Use the reference architectures, shared services, and engineering patterns the team has established. Contribute back when you find a gap, and raise it through the team rather than working around it. Shared patterns are how the team scales. 

• Evaluate before you ship. Design evaluation methodology before the system goes anywhere near production: precision, recall, calibration, drift, business outcome metrics, and A/B tests where they make sense. Measure what matters and operate to it. A handful of promising examples is not evaluation. 

• Own deployment and operations. CI/CD, infrastructure as code, observability, cost. Code does not go over a wall. Production is designed for from the first commit, with enough proximity to the running system to debug it when something goes wrong. 

• Partner for handoff. When the system is ready to move into a vertical, co-build the handoff with the receiving team: documentation, runbooks, on-call posture, and ownership transition. Leave it operable by others, not as a black box tied to one person. 

• Communicate across the altitude range. Translate model and engineering trade-offs for business stakeholders, and explain the same decisions to engineering peers. Both happen regularly and both matter. 

What we are looking for 

• Enterprise product engineering background. Proven experience building and shipping production enterprise systems, with a deep understanding of what production-grade engineering looks like at institutional scale: versioning, testing, code review, release management, deprecation, and operational ownership. The AI work builds on that engineering foundation. 

• Real machine learning and deep learning depth. Hands-on experience building and shipping systems using classical ML (gradient boosting, regression, clustering, tree-based methods, dimensionality reduction) and modern deep learning, with solid grounding in the relevant terminology, mathematics, and evaluation methods. Strong proficiency in Java and Python is required. This role requires at least two years of production ML or DL experience beyond generative AI, as well as familiarity with agentic AI frameworks such as LangGraph or Strands. 

• The right algorithm for the problem. Ability to articulate trade-offs across techniques (cost, latency, reliability, explainability) and select the right approach for the problem, not just the most fashionable one. Model selection is a deliberate, defensible decision. 

• Model evaluation discipline. Experience designing evaluation methodology for production ML systems: selecting appropriate metrics, building offline and online evaluation frameworks, running A/B tests, and applying statistical inference to validate results. Evaluation is how the work is verified, not an afterthought. 

• Infrastructure as code and automation by default. Fluency with Terraform or comparable IaC tooling, container orchestration, CI/CD, and deployment automation. Production systems are deployed through repeatable, automated processes. Observability, cost control, security, and graceful degradation are design inputs from the start, not items addressed at the end. 

• Cloud-agnostic mindset. Comfort designing systems that are not unnecessarily tied to a single cloud provider. Managed services are used deliberately, with lock-in documented. Able to move fluidly across AWS, Azure, and GCP. 

• Altitude range. Able to engage with business leaders to understand a process, align with architects on patterns, and then write the code that does the work. Moving between those modes is a core part of this role. 

Nice to have 

• Familiarity with front-end frameworks (React or comparable) for building end-to-end user-facing AI features. 

• Production experience with vector databases, retrieval systems, or knowledge graphs. 

• Familiarity with MLOps tooling: model registries, feature stores, training pipeline orchestration. 

• Prior experience in research, academic, or mission-driven institutional environments. 

What this role is not 

• A pure prompt-engineering role. LLMs are a constant part of the work, but the job is full-stack AI engineering across the broader ML and DL toolkit. Candidates whose AI experience is limited to the last two years of generative AI will find this role is not the right fit. 

• A data engineering role. This role partners with the data and integration teams on the data and knowledge layer, but does not own the pipelines that feed it. 

• A research role. Following the field and running real experiments is part of the work, but the primary commitment is shipping production systems that meet business outcomes, not publishing or chasing the state of the art. 

• A role for someone earlier in their engineering career. Real production experience across software engineering, ML or DL, and modern cloud-native infrastructure is required. Six or more years in the field is the baseline.