Teach a machine to earn the right to act.
Kā Labs builds the system of record for healthcare's decisions. Frontier models supply the reasoning. We build what makes a decision safe to rely on: the evidence behind it, the rule it applied, the replay that proves it, and the line it never crosses.
Healthcare spends nearly a trillion dollars a year on administration. Much of it goes to checking decisions someone already made.
Every party reviews every other party's work because none of them can rely on the others' judgment. We're building the layer that lets them. Each decision is recorded with its evidence and the policy it applied, replayable years later, and graded against what actually happened. Authority is earned from that record one decision family at a time, and it can be taken back.
Kā is a studio. Proven decision families become standalone companies, all built on one shared substrate. Some of the people who build the substrate will go on to lead the companies built on it. Read the thinking behind it in Decided once, the first entry in The Record.
The problems are unsolved, and the work compounds.
Unsolved, on purpose.
None of our open problems has a textbook answer. When should a machine say it isn't sure, and how do you prove it knows? How do you compile a payment contract into code without losing legal correctness? These are the job.
Work that compounds.
Every decision family we prove makes the next one cheaper to prove. What you build carries into every company the studio launches.
Close to the consequences.
We build inside payer, provider, and risk-bearing environments. Your work meets live cases, and the clinicians who judge them, early.
Small and senior.
Founders who have built and exited in health tech and run risk-bearing care at scale. Nothing sits between you and the decisions that matter.
We run the company on the same rules we build into the product.
-
Trace
Record the decision.
We write down what we knew, what we chose, and why. Memos before meetings. Anyone can replay how we got somewhere.
-
Abstain
Say when you're not sure.
Calibrated abstention is a product primitive and a team habit. "I don't know yet" is a respected answer. Confident guessing isn't.
-
Ladder
Earn the next rung.
Scope here works like our autonomy ladder. Show evidence and you get more authority, quickly. If the evidence turns, it comes back without drama.
-
Ceiling
Some lines don't move.
Patient safety, privacy, and honesty with the people who rely on us. No result is worth crossing them.
-
Record
Argue in public.
We publish our theses in The Record, including the parts that turn out to be wrong. Inside, the best argument wins, whoever makes it.
Each role owns a piece of an open problem.
Founding Engineer, Decision Substrate
Build the layer every Kā decision runs on. The decision model, the evidence, and the replay are preserved by construction, so a decision made today can be replayed and graded years from now.
Open problems · Q03 · Q04
You'd own
- Deterministic replay of model-assisted decisions across model, software, and hardware changes
- The decision trace: inputs, the rule and authority in force, the approving human, the outcome
- Compiling policy and contract logic into executable, auditable artifacts
You might fit if
- You've built systems where correctness and audit matter more than speed: payments, ledgers, compilers, infrastructure
- You care about provenance as much as performance
Platform Build Engineer
Build the platform the substrate runs on: secure, tenant-isolated environments inside payer and provider settings, with builds pinned tightly enough that a decision can be replayed exactly, years later.
Open problems · Q04 · Q05
You'd own
- Reproducible builds and pinned environments, so a change in model, software, or hardware never silently changes a replayed decision
- Tenant-isolated deployments that run inside partner environments with healthcare-grade security and privacy
- Ingestion for claims, clinical, and policy data, with provenance kept from source to decision
- Observability and the audit trail partners and regulators will ask for
You might fit if
- You've built and run production infrastructure where uptime, security, and audit all mattered: healthcare, fintech, or regulated cloud
- You treat infrastructure as code and reproducibility as a requirement
Research Scientist, Calibration and Evaluation
Decide when a machine should say "I'm not sure," and build the evidence that proves it knows. Your metrics decide when a decision family earns a rung, and when it loses one.
Open problems · Q01 · Q05 · Q06
You'd own
- Calibration and abstention methods for clinical and coverage decisions
- Blind-shadow evaluation design, and the gates that promote or demote a decision family
- Evaluation infrastructure that approaches trial-grade evidence at the speed of software
You might fit if
- You've done deep work in calibration, uncertainty, causal inference, or evals
- You'd rather publish an honest error rate than a flattering benchmark
Applied ML Engineer, Clinical Reasoning
Make frontier models reason over policy, clinical evidence, and a patient's history, and make that reasoning something a clinician can inspect and a certifier can grade.
Open problems · Q02 · Q04
You'd own
- Reasoning pipelines that apply the policy in force to the evidence presented
- Interpretability where the audit unit is a clinical trajectory, not a single inference
- Long-horizon context that holds across handoffs and gaps
You might fit if
- You've shipped LLM systems into production where wrong answers cost something
- You read the failure cases before the averages
Clinical Lead
Sit inside the system as its judge. Measure its decisions against outcomes, design the gold sets it's graded on, and own the cases a model should not decide.
Open problems · Q01 · Q06
You'd own
- Clinical review standards and gold-set design for each decision family
- Adjudicating disagreements between the system and expert reviewers
- Drawing the clinical ceiling: what stays with licensed clinicians, whatever the evidence
You might fit if
- You're a practicing or recently practicing physician with utilization review, quality, or informatics experience
- You're comfortable telling engineers "no," with evidence
Founding Operator, Partnerships
Bring payers and providers to the table around a single decision class, and turn shadow evidence into live authority.
Open problems · Q01 · Q05
You'd own
- Partner relationships from first conversation to a decision family running in blind shadow
- The operating cadence that moves evidence through certification
- Carrying what the field teaches us back into what we build
You might fit if
- You've worked inside a payer, provider, or risk-bearing organization and know where decisions get stuck
- You can run a pilot and a negotiation in the same week
Don't see your role? If one of the open problems won't leave you alone, apply anyway.
Open application →Skip the cover letter.
Tell us which problem you'd start on and what you'd try first. A founder reads every application.
Received.
Thank you. A founder reads every application, and if there's a fit you'll hear from us directly.
Kā Labs is an equal opportunity employer. We consider qualified applicants without regard to race, color, religion, sex, sexual orientation, gender identity or expression, national origin, age, disability, veteran status, or any other characteristic protected by law. If you need an accommodation to apply, write to jobs@kalabs.io.