Part-Time Consultant → Key Personnel Option

Location: Remote / US-based (US Citizen or Permanent Resident)

Type: Part-Time Consultant with Key Personnel Option

Compensation (conditional on award):

• $200-300/hr consultant retainer (experience-dependent)
• Option to convert to 0.3 full-time equivalent (FTE) Key Personnel (~$10,/month)

TIMING: This is a post-award role
. We are currently applying for ARPA-H ADVOCATE funding (decision expected Q2 2026). We're building our pipeline now for rapid engagement if awarded.

Regain is applying for the ARPA-H ADVOCATE Program , a $50M+ federal initiative to create the first FDA-authorized autonomous AI for cardiovascular care.

Most teams applying for ADVOCATE are proposing to build. We're demonstrating working software
:

| System | What It Does | Status ||--------|-------------|--------||
Deutsch (cardiovascular agent; Technical Area 1, or TA1) | Autonomous cardiovascular care: three-agent clinical reasoning (Arg Med debate), protocol-bound guideline-directed medical therapy (GDMT) titration, multi-domain treatment plans across heart failure (HF), post-MI, hypertension (HTN), hyperlipidemia (HLD), and AFib | 100% core complete, 4,000+ patients in production ||
Popper (supervisory agent; Technical Area 2, or TA2) | Deterministic safety supervision: 7-gate pipeline gating every clinical output in 47ms. No large language models (LLMs) in the safety-critical path. Versioned policy rules, not neural networks | 100% core deployed ||
Hermes Protocol | Open-source TA1TA2 communication standard (Apache 2.0). Any TA1 works with any TA2. Conformance test suite included | Released Feb 5, 2026 |

Total regulatory budget across both TAs: ~$1.65M over 39 months. This role shapes the FDA strategy for all of it.

You're not filing a 510(k) for a diagnostic app or a remote monitoring tool. You're defining the FDA pathway for autonomous clinical AI that generates medication orders
, a category that doesn't exist in FDA's classification system today.

Here's what makes it unusual:

Deutsch isn't clinical decision support. It generates protocol-bound medication proposals for GDMT titration (angiotensin receptor-neprilysin inhibitors, beta-blockers, mineralocorticoid receptor antagonists, SGLT2 inhibitors, statins, anticoagulants) that route through clinician governance and TA2 safety supervision before reaching an electronic health record (EHR) signature workflow. The intended use statement reads: "autonomous, guideline-concordant cardiovascular medication management and monitoring for adult patients with HF, post-MI, HTN, HLD, and AFib under clinician protocol governance and independent TA2 safety supervision."

No predicate device exists for this.

Popper isn't another AI watching an AI. It's a deterministic policy engine. Same input + same policy pack = same decision, every time. No LLMs, no stochastic outputs, no drift in the safety path. The Medical Device Development Tool (MDDT) qualification would establish a new federal standard for evaluating AI safety monitoring.

Both systems already work. Deutsch manages 4,000+ patients in production. Popper's 7-gate pipeline runs in 47ms. The pre-clinical evaluation framework covers 68 clinical vignettes across 8 cardiovascular conditions with 6 deterministic scoring layers mapping directly to ARPA-H metrics. You'd be defining regulatory strategy for production software, not reviewing a research prototype.

This role requires a dual-track FDA strategy spanning two technical areas on a single program:

| Milestone | Target | Budget ||-----------|--------|--------|| 513(g) Request for Information | Phase 1A, Month 6 | $100K (meeting prep) || Breakthrough Device Designation | Phase 1A, Month 9 | $75K (BDD application) || De Novo regulatory roadmap | Phase 1A | $100K (strategy) || Pre-submission meeting | Phase 1B, Month 18 | Included in Phase 1B || De Novo authorization submission | Phase 2 | $300K (Phase 2 regulatory) |

Key design features you'll need to understand and articulate to FDA:

• Arg Med debate architecture
  : three agents generating, critiquing, and selecting clinical hypotheses using Hard-to-Vary (HTV) epistemic scoring (≥0.7 threshold to proceed,