About the role
Borderless — Fractional CTO / Founding CTO (HITL Orchestration + Decision Infrastructure) Mission Borderless builds the decision-resolution engine for regulated AI in healthcare: an orchestration layer that captures model outputs → human interventions → final attestation → real-world outcomes and turns that loop into repeatable training data, evaluation, and safe automation.
We are not optimizing for “feature breadth” or an EHR UI. We are building the trust + training substrate that makes high-stakes automation measurable, governable, and scalable.
Role summary You will design and ship the core infrastructure that makes Borderless valuable: Canonical Decision Object (CDO): immutable, auditable, replayable, model-agnostic decision records. HITL capture: event-based logging of human edits/approvals/rejections independent of UI.
Outcome binding: deterministic linkage from each decision to downstream outcomes (e.g., claim paid/denied). Model invocation + policy layer: versioned, reproducible model calls with declarative constraints, thresholds, and exclusions.
Evaluation harness: regression tests, golden sets, dashboards, drift/quality monitoring. This is a “platform CTO” role centered on orchestration, provenance, and evaluation—the system that turns messy real-world decisions into a training/evidence flywheel.
What you will own (non-negotiable) CDO v1 spec + implementation (append-only, replayable, auditable) Orchestration framework (routing, queues, retries, idempotency, approvals, policy gates) HITL instrumentation (human deltas as first-class events) Outcome-binding pipeline (link decisions to objective results) Eval + trust substrate (metrics, dashboards, red-team/rollback, provenance) Security + governance-by-design (RBAC, audit logs, encryption, access review) Initial wedge (expected focus) Medical coding → claim submission → payment outcome (starting in outpatient dermatology) because it provides:
• objective/fast feedback, • high economic leverage, • natural human review loop, • measurable ground truth.
Responsibilities
• Architecture & sequencing
Define a staged plan that prioritizes decision→outcome resolution data over product surface area. Make build/buy calls for eventing, storage, workflow engine, observability, and model serving. Establish the “truth model”: what is append-only, what is derived, what is reversible.
• Core platform build (hands-on early)
Implement CDO schema + storage strategy (append-only log + queryable views). Build the orchestration runtime: task routing, HITL queues, retries, idempotency, policy checks. Build model invocation layer: multi-model support, versioning, replay, prompt/config provenance.
Build policy layer: declarative constraints, thresholds, regulatory exclusions, versioned and testable.
• HITL training + evaluation infrastructure
Capture human interventions as structured events: edits, rationale, approvals, rejections. Build evaluation harness: golden datasets, regression suites, error taxonomy, drift monitoring and alerting, “what changed?” diff tooling across model versions/policies. Produce “model readiness” gates for when automation can safely increase.
• Outcome binding (closed-loop)
Design deterministic mapping from decisions to outcomes (e.g., payer adjudication results). Ensure outcome data is linked back to the originating decision record (lineage).
• Security, compliance, and trust
Implement: RBAC, audit trails, encryption, secrets management, environment isolation. Define pilot-ready posture (BAAs, incident response basics, access review cadence).
• Team and execution model
Fractional CTO: set standards, direct contractors/vendors, keep architecture coherent, deliver thin vertical slice. Founding CTO: recruit initial team (platform/backend, integrations, infra/security) and lead execution.
30 / 60 / 90-day deliverables 30 days — “Define the substrate” CDO v1 written spec (fields, invariants, lineage, replay rules). System architecture doc: eventing + storage + orchestration + eval. Repo + CI/CD + environments + baseline observability.
60 days — “Close the loop” Live orchestration path for one workflow (coding-focused): input context → model invocation → HITL review → final attestation Outcome-binding prototype for at least one objective outcome signal (even if partial).
90 days — “Make it repeatable” Eval harness live with golden sets + regression and dashboards. Policy layer versioned and testable; safe rollout/rollback mechanics. Second model or second workflow variant added with minimal incremental architecture work (proof of platform leverage).
What we are looking for
Must-have Built event-driven / workflow / orchestration systems with reliability concerns (retries, idempotency, replay). Deep instincts for data provenance, auditability, and governance (append-only logs, lineage). Experience building evaluation infrastructure (quality metrics, regressions, monitoring/drift).
Ability to scope ruthlessly and ship thin vertical slices. Strongly preferred Experience in regulated domains (healthcare/fintech) with audit trails and access controls. Familiarity with claims/coding/RCM workflows OR willingness to learn quickly with domain experts.
Comfort with multi-model architectures and reproducibility (versioning, deterministic replay where possible).
Working model & comp (stage-dependent) Fractional (8–25 hrs/week): cash retainer + meaningful equity tied to deliverables and time commitment Founding CTO (full-time): founder-level equity + stage-appropriate cash Success metrics (how we’ll judge it) Every decision is replayable, auditable, and outcome-bound.
Human review is captured as structured deltas, not lost in UI. We can prove measurable improvement across model versions with regression discipline. Automation can increase safely because policy gates + HITL + rollback are real, not aspirational.