TheEverydayAIAccelerator exists to turn generative AI into daily reality across HHMI’s administrative and operational functions. This role owns HHMI’s knowledge management layer for AI: the discipline of turning institutional information (documents, records, policies, scientific content, operational data) into structured, retrievable, trustworthy knowledge that AI systems canactually use

The work is technical and grounded. You will design andoperatethe retrieval-augmented generation pipelines that every Accelerator project dependson:the chunking, embedding, indexing, and retrieval patterns that turn HHMI’s content into something AI can find and reason over. For use cases where a graph representation is the right tool (complex entity relationships, lineage, multi-hop reasoning), the knowledge graph gets built andoperatedalongside it. This work happens in partnership with the Principal AI Architect, who owns the AI platform and engineering foundation, and the Technology and Systems Management (TSM) Data Integrations team, who owns the data pipelines from source systems. Whoever holds this role designs theknowledgearchitecture and is accountable foroperatingit.

Why this role matters

HHMI’s scientific, financial, and operational knowledge lives scattered across documents, databases, and systems never built to talk to AI. Without someone accountable for turning that information into something structured and trustworthy, every AI initiative at HHMI either repeats the same expensive groundwork or surfaces answers no one can stand behind. This role solves that problem once so that every Accelerator project and future AI effort can build on a governed, reliable knowledge foundation instead of reinventing it.

What you willactually do

• Own HHMI’s knowledge management architecture.Design how institutional content is captured, structured, classified, retrieved, andmaintainedover time. Makethe callson representation (chunked text, embeddings, structured records, knowledge graphs, or hybrid) for each kind of content and each kind of usecase, andown the consequences.

• Build andoperatethe RAG pipelines.Design and run the retrieval-augmented generation systems that every AI product at HHMI consumes, includingdocument processing, chunking, embedding, indexing, hybrid retrieval, re-ranking, queryrewriting. New projects inherit proven patterns; they do not roll their own.

• Build knowledge graphs where the use case requires it.For problems where graph representation is the right tool (complex entity resolution, multi-hop reasoning, lineage and provenance, relationship-heavy queries), design the data model, stand up the graph store, andoperateit.

• Extract structure from unstructured content.Build the pipelines that turn HHMI’s documents (policies, applications, financial records, scientific content) into something AI can consume. Use the right mix of LLM-based extraction, classical NLP, and rule-based methods for each source, and be able to explain why.

• Solveentityresolution.The same person, fund, application, or concept appears across many systems with many representations. Build the deduplication, linking, and canonicalization that lets the institution rely on a single, defensible truth.

• Govern knowledge classification and lineage.Sit in the AI governance group as the technical voice on knowledge sensitivity, provenance, and retention.

• Partner with Data Integrations and the AI platform team.TSM Data Integrations owns the plumbing across HHMI’s source systems, and you define what AI needs from it while co-building the contracts that connect the two layers. The Principal AI Architect owns the AI platform; the knowledge layer it reasons over comes from this seat.

• Communicate across the altitude range.Translate knowledge-architecture trade-offs for engineering teams, then turn around and explain the same decisions to a business leader or executive in terms thatactually landExpect to do both regularly.

What we are looking for

• Real production RAG experience.Proven experience shippingretrieval-augmented systems and running in production, with failures debugged back through the pipeline and the broken step rebuilt. Hybrid retrieval, chunking strategy, query understanding, and re-ranking used as working tools, not just concepts. This is thecore of therole.

• Knowledge management and data modeling.A librarian’s instinct for content (what’s authoritative, what’s stale, who can see it), plus the ability to look at an unfamiliar domain and identify the right entities, relationships, and representation, defending why an attribute is a node, an edge, or not modeled at all.

• Knowledge graph and entity resolution experience.At least one knowledge graph designed, built, andoperatedin production, with a clear sense of when a graph beats a vector store or document chunk. Deduplication and linking problems solved where the same thing has seven names across foursystemsand none of them are wrong.

• Information extraction and production rigor.Extraction pipelines built to turn unstructured text into structured knowledge using a mix of LLMs, classical NLP, and rule-based methods, treating embedding versioning, retrieval evaluation, corpus drift, and re-indexing as first-class engineering concerns. “The model gave the wrong answer” is adebuggablesystem, not a shrug.

• Data engineering, security, and range.Fluency in adata integration team’s tools(SQL, Databricks,dbt, ETL patterns) pairs withdesigning arounddata classification, access controls, and PII handling from the first conversation rather than as a final review. Ability to lead engineers technically without a reporting line, and to explain the same decision to a non-technical stakeholder in terms that help them choose.

• Technical range tooperateat this level.Strongproficiencyin Python and SQL. Production experience with vector databases (Postgrespgvector, Pinecone,Weaviate,Qdrant, or comparable) and embedding pipelines. Production experience with at least one graph database (Neo4j, AWS Neptune,JanusGraph,TigerGraph,Stardog, or comparable) and graph query languages (Cypher, SPARQL, or Gremlin). Working knowledge of modern NLP and information extraction. Fluency in the modern data stack (Databricks,dbt, or comparable).

• Education and experience.Bachelor’s degree or equivalent, plus at least eight years of hands-on experience across data engineering, information retrieval, and applied machine learning, with at least three years focused on production knowledge management for AI systems (retrieval-augmented generation, knowledge graphs, or both).

Nice to have

• Background in library or information science, formal ontology, or semantic web technologies (RDFS, OWL, SKOS).

• Experience with hybrid retrieval (graph + vector) andGraphRAGpatterns.

• Familiarity with MCP, structured-output patterns, and AI agent tool design.

• Experience with master data management, data catalogs, or lineage tooling at enterprise scale.

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

What this role is not

• A data engineering role. This role partners with TSM’s Data Integrations team on source-system pipelines and the datawarehouse, butdoesn’town that plumbing; it owns the knowledge layer that sits on top of it.

• An AI platform or AI infrastructure role. The Principal AI Architect owns the AI engineering foundation, the platform services, the reference architectures, and the production deployment patterns. Your job is to ensure there is structured, retrievable, trustworthy knowledge for that platform to reason over.

• A pure research role. Staying current on the field (RAG, knowledge representation,GraphRAG, neuro-symbolic methods) matters, but the work is building and operating production knowledge systems, not publishing about them.

• A role for someone whose RAG or graph experience is only academic or prototype-scale. You need production scars: systems that have handled real users, real failure modes, and real corpus evolution over time.

Practical details

This role is hybrid, with 3 days per week in-person at HHMI's offices in Chevy Chase, MD. It reports to the Director, AI Enablement.

We encourage qualified candidates who are eligible to work in the United States to apply. Please note, we are not able to sponsor a visa for this position at this time 

#LI-EG1