Frequently asked questions.

Each cycle is the first in a series, not a one-time engagement. The first cycle produces a working system and proves the model. Subsequent cycles expand scope: more workflows, more data sources, more regulatory domains. Most clients run three to six cycles over six to twelve months.

We build the deterministic layer first: a governed data pipeline and a rules engine that produce validated, auditable outcomes. The AI layer sits on top and explains what the system already determined, with a citation for every answer. That is the difference between an AI demo and a system someone can sign off on.

It depends on the workflow, so we scope it first. In a 30-minute call we ask what the workflow is, what data sits behind it and what done looks like. If it fits a four-week cycle, we tell you exactly what it includes and what it costs. If it does not, we tell you that too.

Yes. We are based in San Diego and have deep relationships in the Southern California life-sciences ecosystem, but we deliver remotely for engagements across the US.

Yes. We work in isolated environments, execute BAAs before any data is accessed, and tokenize PHI at the ingestion boundary so protected data never reaches the AI layer directly. More on security & PHI →

You do: the code, the infrastructure definitions and the documentation. Everything is deployed into your cloud account using open-source and platform tooling your team can maintain independently.

GRC platforms are systems of record for compliance documentation: they store your policies, track your attestations and manage your audit calendar. This evaluates operational events against those requirements in real time, before records are committed. Different layer. Most clients keep their GRC platform and add this underneath it.

No. We codify the requirements your compliance team defines and validates. Your compliance officer owns the interpretation; we make that interpretation executable, consistent and auditable. The system cites authority. It does not become authority.

Rules are versioned. When a requirement changes, the affected rules are updated and the change is timestamped, so you can demonstrate exactly which version of a requirement was in force when a given event was evaluated. Ongoing rule monitoring and updates are available as a managed operations engagement after the first cycle.

Structurally, no, and we test for it. The copilot can only reference rule keys that exist in the engine catalog and citations attached to those rules. In the Luveo evaluation across 30 scenarios we recorded zero hallucinated rule keys and zero fabricated regulatory citations. When the system lacks grounding for a case, it says so rather than generating a plausible answer.

You probably could. The question is opportunity cost: every sprint your team spends on ingestion and transformation infrastructure is a sprint not spent on product. We have built this pattern repeatedly, so we are not learning on your timeline. And at the end of the cycle it is your code in your environment, so there is no long-term dependency.

No. We connect to whatever you already use: Tableau, Looker, Power BI, Metabase. This is the layer underneath your dashboards, not a replacement for them. If your visualization layer is fine and your data is not, this is the right fix.

It complements it. AutoML platforms are good at what they do, but a model can only be as accurate as the data feeding it. If there are structural gaps in collection or inconsistencies across sources, the tool cannot detect that; it will train confidently on incomplete data. This fixes the input problem underneath.

You do: the code, the infrastructure definitions and the documentation. Everything is deployed into your cloud account using open-source and platform tooling your team can maintain independently.

No, and that is by design. The system does not diagnose, recommend treatment or analyze physiological signals. It surfaces reference material from your own documentation with a citation and requires a qualified clinician to review before acting. We document this positioning at engagement start and design to hold it. If a use case you want would cross into SaMD territory, we will tell you that rather than build it.

Three structural differences. The corpus is bounded and approved, so the system cannot draw on model training data. Every response carries verifiable provenance, and unsourced output is a failure condition rather than a normal one. And refusal behavior is tested: we build an evaluation set including out-of-scope and ambiguous cases, and your clinician scores whether the system declined when it should have.

Every response cites the specific source document and section, so a clinician can verify against the original immediately; that is the primary control. Beyond that, the week-three evaluation is scored by your clinical reviewer before the system reaches staff, and the corpus is versioned so you can see exactly which document version produced a given answer. The system is reference material subject to clinical review, and it says so in every response.

The corpus is versioned. When a protocol is updated, the new version is ingested and the prior version is retained for audit, so you can demonstrate which guidance was in force at a given time. Ongoing corpus maintenance is available as a managed operations engagement after the first cycle.

No. We do not build molecular property prediction models or do de novo generative design, and we do not have a proprietary chemistry or biology platform. We build the governed data and knowledge infrastructure those efforts depend on.

One bounded workflow in a four-week cycle: one assay family, one program’s documentation, or one decision workflow. Scope stays narrow enough that your scientists can validate the output against data they know well.

You do: code, infrastructure definitions, documentation and any data model we build, deployed into your environment on tooling your team can maintain without us.

Research engagements require domain immersion we cannot do at volume. We would rather take fewer and do them properly, and tell you early if the fit is not right.

If your state expanded Medicaid, yes: the six-month redetermination mandate is federal and cannot be waived. If your state did not expand, that specific provision does not apply to your panel, though managed-care documentation and denial exposure usually still do. We establish this in the first ten minutes of a scoping call.

Your billing company submits claims. This evaluates whether a claim will survive submission before it goes out, and whether a patient’s record supports continued eligibility before their renewal date. Most billing partners welcome it: fewer denials to rework is good for both of you.

Documentation quality and documentation compliance are different problems. A thorough note can still fail a plan-specific modifier requirement. The question is not whether your clinicians document well; it is whether anything systematically checks that documentation against the requirement for that specific claim, payer and program.

That is exactly why the cycle is four weeks. A cycle started this quarter is in production well before the first six-month renewal wave. A cycle started in November is not.

Partially, and not reliably. H.R. 1 allocated $200 million in FY 2026 to help states modify eligibility and enrollment systems, far less than states will actually spend, and federal guidance on those changes has been slow to arrive. Assuming the state will catch your documentation gaps is a risk, not a plan.