Open position · Knowledge Engineering
Before an agent can practice medicine, navigate contract law, or manage a derivatives book, someone has to pour the right knowledge into it. The vocabulary, the reasoning patterns, the unstated assumptions that practitioners carry in their heads. That someone is you.
The role
A general-purpose agent is a blank slate. A Domain Imprinter turns it into something that a cardiologist, a derivatives trader, or a civil engineer would trust. You are the conduit between a field of human expertise and the agents that operate within it — extracting tacit knowledge from practitioners, translating it into the structures agents can absorb, and validating the result against the domain standard you know from the inside.
This is not prompt engineering in the hobbyist sense. It's the work of making an agent genuinely fluent — aware of the exceptions, the edge cases, the professional norms, and the liability landscape of a real industry. When an imprinted agent fails, it usually fails because the imprint was shallow. Your job is to make it deep.
What you'll do
Knowledge extraction
Conduct structured interviews with domain experts to surface tacit knowledge that practitioners rarely articulate
Map the vocabulary, mental models, and decision heuristics that define expert judgment in your field
Identify unstated assumptions — the things every practitioner knows but no textbook says
Document edge cases, jurisdiction-specific rules, and context-dependencies that separate expertise from surface knowledge
Agent configuration
Translate extracted knowledge into system prompts, context documents, and structured knowledge bases
Configure agent personas that are credibly fluent — not just accurate, but professionally appropriate
Tune agent risk tolerance, communication style, and escalation thresholds to match domain norms
Build RAG corpora from authoritative domain sources and validate retrieval quality
Validation & testing
Design domain-specific evaluation suites — questions only a genuine expert would answer correctly
Test imprinted agents against practitioners and close the gaps they identify
Flag areas where agent knowledge is shallow, outdated, or confidently wrong
Refresh domain imprints as the field evolves — regulations change, best practices shift
Cross-functional work
Partner with Skill Architects to layer procedural skills onto the domain knowledge foundation you build
Brief Bot-Psychoanalysts when domain confusion appears to be causing behavioral drift
Present imprinting plans and fidelity reports to clients before agent deployment
Domains we're imprinting — pick at least one
Clinical medicine
Diagnosis, protocols, liability, triage
Legal & compliance
Contracts, jurisdiction, regulatory risk
Financial services
Trading, credit, reporting, audit
Engineering
Standards, failure modes, tolerances
Scientific research
Methodology, peer review, reproducibility
Policy & regulation
Rulemaking, interpretation, enforcement
What we're looking for
Must-haves
7+ years of practitioner-level experience in your chosen domain — you've done the actual work
Ability to articulate implicit knowledge: you can explain what you know and why, not just what to do
Intuition for how AI systems absorb and misrepresent expertise — you've seen it fail
Strong synthesis skills — you extract signal from hours of expert interviews and mountains of documentation
Comfort with the limits of encoding: some knowledge resists imprinting, and you know when to stop
Nice-to-haves
Second domain specialization — cross-domain imprinters are in high demand
Background in knowledge engineering, ontology design, or information architecture
Experience building RAG pipelines, fine-tuning datasets, or evaluation frameworks
Consulting or advisory background — you're used to translating deep expertise into structured deliverables
Technical baseline
You are a domain expert first. The technical layer is learnable — and we'll support you in learning it. What matters is that you develop enough fluency to build and test knowledge bases without engineering hand-holding, and enough intuition to know when an agent's output reflects the domain you gave it versus something it hallucinated around it.