Jobs

Principal AI Architect
Function: AI / ML
Reports to: Chief AI Officer
Level: Principal IC
Location: Remote (U.S.)
Scenario: Platform Extension
Target start: Search timeline 6–9 months; start as soon as the right hire is found
About Via Health
Via Health is building a real-time operating system for hospitals that connects fragmented clinical and financial data (EHR, supply chain, claims, staffing, and more) into one actionable intelligence layer. Via uses AI to form clinically meaningful patient and procedure groupings, identify unwarranted variation, and surface "next best actions" that reduce costs while improving outcomes—so health system leaders can move from dashboards to measurable change and accountability. We partner closely with leading health systems as design partners to validate the platform in production and scale what works across the broader hospital ecosystem.
About the role
Our current platform—running in production on Databricks with three health system design partners—already does three hard things at once. Clinical Topic Extraction (CTE) runs LLM agents that produce structured, evidence-backed outputs from operative reports. CliniGraph consumes those extractions to build patient-similarity graphs and uses community detection plus LLM matching to align clinical cohorts to Procedure Care Groups. Synthetic Data Generation produces clinically coherent synthetic datasets with closed-loop validation against CTE.
The next phase is bigger than what's shipped. We are extending the graph family beyond procedures into diagnoses and claims, building a set of insight agents on top of the new graphs (clinical supply, patient outcomes, cost accounting, total cost of care), maturing the MLOps layer, and running a structured refinement program on the agents themselves. That is not a one-person workload.
We are hiring a Principal AI Architect to partner with the Chief AI Officer on hands-on build of every major AI initiative in this next phase. This is a senior IC role—you will write code, design systems, write the ADRs, and own implementation across graph construction, LLM agents, evaluation harnesses, and the operational stack that makes those agents safe to put in front of clinical and financial users.
What you'll do
• Partner with the Chief AI Officer on the full AI roadmap: extending the graph family, shipping insight agents, maturing MLOps, and refining the agents in production.
• Design and build the next generation of clinical graphs: additional procedure-based graphs beyond the current care group families, diagnosis-based graphs anchored on disease trajectories rather than surgical episodes, and claims-based graphs for payer-side and total-cost-of-care use cases. Decide where these share infrastructure versus where they need to diverge, and write the ADRs that govern those boundaries.
• Build the insight-agent family on top of the graphs: clinical supply analysis, patient outcomes, cost-accounting drivers from Strata, and total cost of care decomposed from claims. These are evaluated, registered, versioned agents with structured outputs and evidence tracking—not chatbots.
• Lead the agent refinement program: systematic prompt and few-shot optimization across care group families, improvements to evidence quote matching, per-family architecture decisions, and prompt versioning tied to ground-truth and LLM-judge metrics.
• Build the AI Agent operational stack: conversation monitoring and logging, automated testing and regression suites, guardrails (input/output filtering, scope enforcement), hallucination detection and review workflows, and cost monitoring at the token and session level. Lead the build-vs-buy evaluation for AI observability tooling (LangSmith, Langfuse, Helicone, Arize Phoenix) and integrate the chosen platform.
• Unify the MLOps layer across all AI workloads: consistent model registration and promotion paths, automated evaluation harnesses gated on care-group-family metrics, drift detection against the synthetic-data closed loop, and reproducible pipeline runs across local dev, staging, and production.
• Establish model governance: versioning of deployed models, prompts, and parameters; output traceability for clinical and financial recommendations; and change-management gates for AI updates once live users are on the system.
• Stand up the clinical validation process: ground-truth evaluation against clinician-reviewed cases, with an evidence trail substantiating savings-with-maintained-outcomes claims for value-based pricing.
• Be a strong technical voice in cross-functional decisions—particularly around the Semantic Layer, the claims data contract, and grounding insight agents against stable schemas.
What we're looking for
• Significant hands-on experience building production LLM and agent systems—prompting, structured outputs, RAG, tool use, evaluation, and agent orchestration—at meaningful scale.
• Graph ML in production: graph construction, similarity projection, community detection, and/or GNNs.
• Direct experience standing up AI observability and the operational stack: monitoring, automated testing, guardrails, hallucination detection, cost controls.
• Senior IC track record. You set technical direction by writing the code and the design docs, not by delegating.
• Healthcare data fluency. You have shipped against real EHR, claims, or cost-accounting data and you know what PHI safety means in practice—familiarity with HIPAA, BAAs, and the realities of PHI handling. You will not learn this on the job.
• Databricks + Unity Catalog at scale: PySpark, Delta, MLflow model registry, Databricks Jobs, model serving endpoints.
• Strong software engineering fundamentals (Python required; comfort across data and application stacks). You write ADRs, run design reviews, and set conventions across repos.
Bonus points
• Direct experience with EHR data (Epic, Cerner, health system EDW), claims data (837/835, episode grouping, risk adjustment), and cost accounting (Strata, EPSi, or comparable).
• Background in clinical AI or healthcare ML, with familiarity with clinical validation frameworks.
• Experience designing model governance for clinical-adjacent decision support.
• Experience grounding agents against semantic layers (Databricks Unity Catalog Metrics, dbt Semantic Layer, Cube, or comparable).
• Experience with synthetic data programs for healthcare—profile-driven generation, closed-loop validation, PHI suppression.
• Experience with prompt optimization frameworks (DSPy, TextGrad, Instructor) in production.
What success looks like in your first six months
• The diagnosis-graph and claims-graph extensions have accepted architectures, with one landed end-to-end against a target cohort.
• The first insight agent (likely Clinical Supply Analysis) is in production with a registered evaluation harness, and a second is in active development.
• An AI observability platform is selected, integrated, and used for monitoring and evaluation in production.
• The AI Agent operational stack (logging, testing, guardrails, hallucination review, cost monitoring) is in place before any insight agent is exposed to design-partner users.
• A unified MLOps baseline is documented and operating: consistent registration, evaluation, drift detection, and cost observability.
• Model governance is documented and operating.
• A clinical validation process is running with clinician-reviewed ground truth.
• The single-point-of-failure risk in AI is resolved—you and the Chief AI Officer can each cover the core stack.
Why Via
• Direct partnership with the Chief AI Officer on every major AI decision in the company.
• The AI work is genuinely consequential: the graphs you build will define how clinical cohorts are reasoned about, and the insight agents you ship will influence surgical, supply, and total-cost-of-care decisions in real health systems.
• A platform with real production substrate to extend—not a greenfield where you are alone at the whiteboard.
• Strong AI-native engineering culture and the budget to do the operational stack right rather than fast.

Via Health is an equal opportunity employer. We evaluate qualified applicants without regard to race, color, religion, sex, sexual orientation, gender identity, national origin, disability, veteran status, or any other protected characteristic.