Director, Research AI

What matters most

• Embed AI in day-to-day research

  The posting identifies this as the primary focus: partner with Research and assay teams, prioritize opportunities in target and ligand discovery and related scientific workflows, and measure adoption and impact.

• Deploy scientifically grounded agentic workflows

  Architect multi-agent systems for human-genetics-based target discovery, siRNA ligand discovery, literature and competitive intelligence, experimental design, data interpretation, and knowledge synthesis. Integrate coding and research agents with evaluation and lab feedback.

• Set the Research AI roadmap and lead the team

  Define use cases with scientific and operational impact, align the roadmap with the broader Data & AI strategy, and recruit and develop a 3–5 person team while providing hands-on technical leadership.

• Integrate securely with enterprise platforms

  Partner with IT to connect AI workflows to enterprise data and lakehouse environments across AWS and GCP, consistent with data governance, security, and regulatory standards.

• Make systems reliable in production

  Oversee pipelines, deployment, evaluation, MLOps, reproducibility, and model validation. The remit also includes computational support for ligand and antibody design and collaboration with the Alnylam Human Genetics team.

Likely challenges to think through

• Choosing depth over breadth

  Preparation hypothesis: the listed use cases span discovery, design, analysis, and intelligence, while the initial team is small. Prepare a way to rank opportunities by scientific value, feasibility, and measurable adoption rather than pursuing every workflow at once.

• Balancing agent autonomy with scientific validity

  Preparation hypothesis: automated research workflows may accelerate work but produce conclusions that do not hold up experimentally. Be ready to explain evaluation criteria, traceability, human review, and how lab feedback would trigger revisions.

• Connecting data without weakening controls

  Preparation hypothesis: synthesizing public and internal knowledge across enterprise platforms could create tension between useful data access and governance, security, and reproducibility. Prepare an architecture discussion that makes access controls, provenance, and validation explicit.

• Combining hands-on delivery with leadership

  Preparation hypothesis: directing the roadmap, building a team, and contributing technically could compete for time. Prepare to explain how you would delegate implementation, set engineering standards, and stay involved in consequential design decisions.

Topics to prepare

• Scientific agent architecture

  The posting calls for multi-agent workflows combining coding and research agents with evaluation frameworks. Revise by diagramming an agent workflow for literature synthesis and target scouting, including handoffs, evidence checks and points for scientist review.

• Human-genetics target discovery

  The role explicitly covers target discovery using human genetics datasets and work with the Alnylam Human Genetics team. Revise by outlining how genetic evidence could inform target prioritization, then identify uncertainties that require domain-expert review.

• siRNA ligand discovery and experimental feedback

  The posting names ligand discovery for siRNA delivery to novel tissues and cell types, plus validation through lab feedback. Revise by tracing a proposed ligand-design workflow from hypothesis to assay result and explaining how conflicting results would change the next iteration.

• Scientific evaluation and reproducibility

  The director is expected to oversee evaluation systems, model validation and reproducible production pipelines. Revise by defining separate checks for scientific grounding, workflow reliability and measurable research impact in an agent-assisted analysis.

• Research AI prioritization and governance

  The role sets a Research AI roadmap, drives adoption and partners with IT on security, data governance and regulatory alignment. Revise by comparing two stated use cases on scientific value, feasibility, adoption and data-access constraints; this comparison is a preparation exercise, not a stated company process.

Skills to revise

• Prototype a Python agent workflow

  Python and agentic or multi-agent development are explicit qualifications. Practice building a small Python workflow in which research and coding agents exchange a structured result, record their inputs and flag unsupported conclusions.

• Build an evaluation set

  The posting requires evaluation frameworks and lab-informed improvement. Practice creating a small set of scientific questions with reference evidence, then score an agent workflow for factual support and errors; describe how assay feedback would update the evaluation.

• Integrate scientific data

  The role involves knowledge synthesis from public and in-house data and large-scale data integration. Practice specifying a data-ingestion pipeline that preserves provenance, distinguishes access levels and produces traceable inputs for an analysis.

• Review cloud deployment design

  The posting names AWS, including Bedrock, and/or GCP, plus integration with an enterprise data platform and lakehouse. Sketch a secure deployment design using a cloud environment named in the posting, and walk through access control, monitoring, reproducibility and failure handling.

• Present a team delivery plan

  The director will lead a 3–5 person interdisciplinary team and partner with Research, assay teams and IT. Practice presenting a short plan for one stated use case, assigning scientific and engineering responsibilities, defining a validation milestone and specifying how adoption and impact would be measured.

• Select a relevant project walkthrough

  Prepare a concise account of an AI/ML system delivered in biopharma research or drug discovery: the scientific problem, experimental data, your technical and leadership decisions, deployment, and measured outcome. Use only work you can discuss.

• Sketch a research-agent design

  As a preparation exercise, diagram a multi-agent workflow for human-genetics-based target discovery or siRNA ligand discovery. Show data inputs, research and coding agents, human review, evaluation, and where laboratory feedback would change the system.

• Draft a use-case prioritization framework

  Prepare a one-page proposal for comparing opportunities in target or ligand discovery, experimental design, and data analysis. Include scientific value, feasibility, assay-team involvement, adoption measures, and evidence of impact; present it as a proposal, not Alnylam’s existing roadmap.

• Assemble production and platform evidence

  Choose examples that demonstrate Python, production AI/ML, LLM integration, and AWS (including Bedrock) and/or GCP experience. Be ready to explain pipeline reliability, reproducibility, model validation, and how you handled data access and security.

• Prepare a cross-functional leadership example

  Outline how you led a technical team and partnered with scientists and IT to ship an AI product or platform. Distinguish your hands-on contributions from team contributions and explain how you resolved scientific or engineering trade-offs.

• Form questions about scientific validation

  Ask how Research, assay teams, and the Alnylam Human Genetics team currently assess AI-supported conclusions, return lab feedback, and measure adoption and scientific impact.

• Form questions about the delivery environment

  Ask which research workflows are highest priority, what enterprise data platform and lakehouse integrations are in scope, and how the director’s planned 3–5 person team will work with Research and IT.