Why We Exist

AI is now trusted with decisions where a missed call costs a life or breaks an audit. Hype isn't a strategy for those decisions. Determinism is — and we build accordingly.

LLMs are excellent at fluency. On their own, they are not good enough for safety-critical calls — crisis detection in a nursing home, compliance enforcement in a regulated codebase, PHI handling, audit-defensible configuration.

Praglogic's posture is to reach for determinism — explicit rules, reproducible outputs, auditable pathways — wherever failure is unacceptable, and to use LLMs where they genuinely help. That posture runs through Lilo Solace's crisis-detection pipeline and EmbedIQ's deterministic configuration generation alike.

Core Principles

Determinism where it matters

Explicit rules, reproducible outputs, auditable pathways. We reach for determinism in the places where probabilistic answers aren't good enough and let LLMs assist with fluency where they genuinely help.

  • Lilo Solace: four-gate deterministic crisis-detection pipeline
  • EmbedIQ: byte-for-byte reproducible configuration generation, no runtime LLM
  • Shared principle: every safety-critical decision traceable to the rule that produced it
Honesty about stage

We ship what's shipping. We flag what's planned. We don't call pilots clinical outcomes or roadmap product reality. Clear stage badges, honest caveats, and published methods are features, not footnotes.

  • Lilo Solace: labelled pre-pilot; FDA De Novo submission targeted Q4 2026
  • EmbedIQ: labelled shipping v2.1 under MIT; roadmap items listed separately
  • Internal benchmark numbers always distinguished from clinical outcomes
Human judgment in the loop, not behind the curtain

LLMs help with fluency, phrasing, and user experience. Deterministic pipelines run the safety-critical path. Humans approve the outcomes that matter — clinical escalation, configuration pull requests, compliance sign-off — so the "AI decision" is never the last word.

  • Lilo Solace: crisis alerts escalate to care-team review, not autonomous action
  • EmbedIQ: configurations land in a pull request for team approval before merge
  • No black-box AI for safety calls; no automation-bias traps

The Centaur's Mind

In Greek mythology the centaur combined human intelligence with equine strength — two disparate natures fused into something greater than either alone. That's the pattern we aim for: human judgment plus machine strength, applied across domains.

How the pattern shows up in our products

Human Judgment

Clinical escalation decisions, PR review for configurations, compliance sign-off, regulatory strategy. The calls that live with consequences.

Machine Strength

Deterministic crisis detection in Lilo Solace. Deterministic configuration generation in EmbedIQ. Continuous monitoring, reproducible outputs, audit trails at scale.

Same pattern, different tools

Clinicians plus deterministic safety for Lilo. Engineers plus deterministic configuration for EmbedIQ. One principle applied across two very different domains.

Team

Our team combines 29 years of enterprise healthcare architecture experience with peer-reviewed clinical AI research.

Enterprise healthcare architecture

Twenty-nine years delivering production healthcare platforms at scale, including fifteen years at a major US health insurer operating claims adjudication, benefits administration, and provider-network systems for more than eleven million members, and ten years at an enterprise software lab on platform architecture.

Clinical AI research

Active research in deterministic safety architectures for clinical AI. Manuscript on the Lilo Engine safety pipeline available as a medRxiv preprint; a second manuscript on architectural safety guarantees for clinical-AI policy is in the publication pipeline. Grounded in published geriatric mental-health literature (behavioral activation, reminiscence therapy, grounding, C-SSRS).

Deterministic AI for regulated tooling

EmbedIQ is the developer-tool expression of the same thesis — an adaptive Q&A wizard that generates byte-for-byte reproducible Claude Code configurations, with built-in compliance packs for HIPAA, PCI-DSS, and FERPA. No runtime LLM, zero data persistence, air-gap compatible — the architectural properties regulated industries need and LLM-based competitors structurally can't match.

Cloud, compliance, and engineering

Certifications across GCP Professional Cloud Architect, AWS Solutions Architect, TOGAF, HIPAA Security Officer, HL7 FHIR, and ITIL 4. Engineering across Python, Go, TypeScript, and Kubernetes, with HIPAA-aligned controls baked in from the start rather than bolted on after.

Current team role

Principal Architect & Research Lead

Responsible for the Lilo Solace safety architecture, the EmbedIQ configuration pipeline, the clinical-instrument framework, and the peer-review and regulatory strategy. Leads engineering, research, and clinical-validation planning across both product lines.

Clinical advisory

We are building a clinical advisory board to guide product development, validation protocols, and regulatory strategy. If you are a clinician, researcher, or policy expert interested in advancing safe therapeutic AI for seniors, we welcome the conversation.

Get in touch →

Our Direction

Mission

Build deterministic AI where determinism matters, and ship honestly about what's true today.

Our Focus:

  • Therapeutic AI for elderly mental health with architectural safety guarantees
  • Compliance-aware configuration tooling that is air-gap compatible and byte-for-byte reproducible
  • Research and regulatory pathways that the work can survive — peer review, IRB, FDA De Novo
  • Product claims that match what ships; roadmap items labelled as such
Vision

A future where AI in regulated domains is trusted because its decisions are reproducible, auditable, and honest about what they are — not because anyone was asked to trust them.

Impact Goals:

  • Deterministic safety pipelines recognised as a distinct class of clinical AI
  • Enterprise AI coding agents configured by audit-defensible tooling, not hand-written CLAUDE.md files
  • Published methods and open specifications that other teams can build on
  • A bar of honesty in AI marketing that the rest of the industry gets dragged toward

Our Values

The principles that guide every decision, from product design to partnership strategy.

Evidence and honesty

We publish methods, document invariants, and caveat what we haven't proven. Internal benchmark numbers are never presented as clinical outcomes. Pre-pilot is labelled pre-pilot. Roadmap is labelled roadmap. The published work stands or falls on its own terms — not on marketing.

Human-Centered Design

Technology should meet humans where they are. Voice-first interfaces that adapt to an 82-year-old resident's preferences for Lilo Solace. Role-adaptive configuration that produces meaningfully different output for a developer versus a business analyst versus a compliance officer for EmbedIQ. Different users, same posture.

Privacy & Security

Regulated data deserves architectural guarantees, not promises. Lilo Solace is HIPAA-compliant by design — end-to-end encryption, seven-year immutable audit logs, role-based access, and an on-premise deployment path so PHI never has to leave the facility. EmbedIQ runs locally with zero data persistence and zero network calls — air-gap compatible by construction.

Ethical AI

No black-box AI for safety calls. Deterministic pipelines run the crisis-detection gate and the compliance-enforcement path; LLMs assist with fluency where they genuinely help. Every safety-critical decision is traceable to the rule that produced it, and humans approve the outcomes that matter — clinical escalation, configuration pull requests, compliance sign-off.

Continuous Innovation

The pursuit of better outcomes never stops. We expand the Lilo Solace clinical-instrument framework, refine the deterministic safety pipeline, and evolve EmbedIQ's generators, domain packs, and skills in response to real enterprise use. Improvements ship with explicit version labels, not silently.

Respect for the humans on the other end

Every product choice gets measured against the person on the other end — the resident receiving a daily check-in, the family reading a mood pattern, the engineer who needs a CLAUDE.md that makes Monday easier, the compliance officer whose audit has to survive. Designing for dignity, respect, and the actual workflow is not a UX afterthought; it is the work.

Join us in building AI that earns the trust it claims.

Looking for collaborators and community members — clinicians, enterprise engineering teams, researchers, advisors, and investors — who share the posture.