Translational AI Engineer

Job description

ROLE SUMMARY: The technical engineering counterpart to AIDE’s applied workflow roles, embedded within the Research Unit to convert promising AI workflow concepts into durable, evaluated, and supportable systems that accelerate translational science. This position is focused less on discovering use cases and more on the gap that appears once a prototype needs to scale in a scientific setting. The role translates recurring needs from computational biology, immunology, and clinical teams into fit-for-purpose AI systems, including generative AI, agentic workflows, predictive models, foundation models, retrieval-augmented systems, and fine-tuned model architectures. It also builds the data, evaluation, integration, and deployment layers beneath those systems, ensuring that internal AI tools do not remain fragile scripts or demos. The ideal candidate is a hands-on translational AI engineer who can move fluidly between scientific intent, model behavior, and production-quality infrastructure. They should be comfortable reading and hardening AI-assisted codebases, standing up ETL (Extract, Transform, and Load) and database foundations, implementing cloud deployment and CI/CD patterns, and building evaluation harnesses that expose scientific, clinical, and technical failure modes before tools move from alpha to beta. The distinguishing strength is translational engineering judgment: turning ambiguous scientific needs into reliable AI systems, tracing failures to their root cause, defining fit-for-purpose evaluation with scientific and clinical partners, and knowing when to build internally versus adopt a commercial AI tool. This is a role for someone who turns promising AI workflows into systems that scientific and clinical partners can trust, reuse, and improve. ROLE RESPONSIBILITIES: Design and build fit-for-purpose AI/ML systems for recurring scientific and clinical workflows, including hybrid RAG when grounding and domain context are required. Take AI-assisted prototypes built across AIDE and Systems Immunology, identify their technical and scientific failure modes, and rebuild parts that need engineering before scaling for wider use. Stand up lean data platform and ETL underneath these systems, in partnership with Digital, so tools do not “rot” into one-off scripts. Own promotion of internal AI tools from alpha to beta, define what “production-ready” means for a given tool, and build the evaluation harness that proves it. Establish and document the SOPs that help others build rigorously from day one, instead of discovering the gap after a tool is already widely used. Rigorously evaluate commercial AI/ML and GenAI tools and vendors against a build-vs-buy bar, covering capability, cost, security, scientific fit, and workflow readiness to support go/no-go purchasing decisions. Run evaluation loops that measure system quality against workflow-specific scientific or clinical benchmarks, using input from computational biologists, immunologists, biologists, and clinicians to drive model selection and iteration. Distill what is learned into hardened primitives, reference architectures, and benchmark harnesses that scale across AIDE’s other tools and deployments. Apply the same rigor to AI that any scientific or translational method would receive: fit-for-purpose evaluation, grounded outputs, documentation, guardrails, disclosure of model limitations, and human oversight. Stay current on translational AI methods and infrastructure, including generative AI, agentic systems, predictive modeling, biological foundation models, retrieval and fine-tuning methods, and fit-for-purpose evaluation practices as they evolve. Contribute to a culture in AIDE where rigor is expected rather than exceptional, including calling out your own failures before someone else must. Build external presence by attending relevant conferences, publishing methods and evaluation results, and presenting materials that showcase the group’s technical approach and impact. BASIC QUALIFICATIONS: PhD with 1+ years of experience OR Master’s degree with 5+ years of software or ML engineering experience - OR Bachelors degree and 6 + years of experience and/or equivalent demonstrated experience shipping AI/ML systems for scientific applications. Hands-on experience building AI/ML systems beyond prompt-wrapping, with depth in at least one modality such as generative AI, agentic workflows, predictive models, foundation models, hybrid RAG, or fine-tuned architectures. Demonstrated ability to evaluate, debug, and harden AI-assisted (“vibe-coded”) codebases: read a prototype someone else wrote quickly with AI assistance, find its failure modes, and rebuild the parts that need real engineering. Strong Python engineer fluent in modern AI/ML tools, including model APIs, prompt engineering, hybrid RAG, fine-tuning, predictive modeling, evaluation tooling, and frameworks such as PyTorch, HuggingFace, LangChain, or LlamaIndex. In-depth database and ETL experience (Postgres or equivalent) - able to stand up the data platform underneath a system, not only call an API. Fluency in cloud infrastructure and deployment: hands-on with AWS, GCP, or Azure, containerization, and CI/CD. Sufficient immunology, biology, translational research, or clinical workflow literacy to understand the problems computational biologists, immunologists, biologists, and clinicians are trying to address. Demonstrated build-vs-buy judgment: has recommended adopting a commercial tool over an internal build where that was the better answer and can explain the tradeoffs that drove the decision. Demonstrated practice of building evaluation harnesses, error analysis, and hardening into systems from the start, with specific examples tied to scientific or clinical workflow requirements. Experience delivering in environments where requirements shifted as the work was learned, including examples of deliberately slowing down when speed was about to become a production liability. PREFERRED QUAL