Skip to evidenceSelected implementation evidence / August 6, 2026
Architecture you can
question and explain.
This page is the public, sanitized engineering record. It shows system boundaries, data flow, test categories, known gaps, and ownership without exposing credentials, private infrastructure, or security-sensitive implementation details.
01 / Flagship systems
Evidence by product.
Care Navigator AI
- Architecture trace
- User task → bounded server route → official health source → schema validation → source-labeled result → human verification
- Inspectable evidence
- Health search, PDF/image extraction, medication identity, FDA label sections, cost-match classification, emergency copy, accessibility, and no-clinical-validation boundaries.
- Release-test categories
- Route schemas, parsers, safety copy, source contracts, cost matching, document intake, rendering, TypeScript, and production build.
Open technical record ↗CareSignal
- Architecture trace
- Organization query → parallel CMS/GLEIF search → human entity selection → five deterministic factors → completeness → local CRM preview
- Inspectable evidence
- No patient or PHI input, visible 0–100 weights, unknown headcount, source health, ambiguity handling, and no-transmission export.
- Release-test categories
- Identity collisions, ZIP semantics, provider outages, score thresholds, provenance, rendering, and boundary regression checks.
Open technical record ↗SparkTutor
- Architecture trace
- Grade + subject + chapter → deterministic challenge bank → normalized answer check → hint ladder → path-scoped mastery → local persistence
- Inspectable evidence
- Versioned learner state, age-appropriate coverage, off-topic coaching boundaries, transparent adaptivity, reset controls, and no runtime-AI claim.
- Release-test categories
- Equivalent answers, 208 scripted chapter interactions, progression rules, persistence recovery, coach boundaries, accessibility, and production build.
Open technical record ↗02 / Ownership
AI-assisted engineering,
human-owned decisions.
I owned
Problem framing, scope, source selection, product and accessibility decisions, implementation review, test acceptance, claims, and deployment approval.
AI assisted
Drafting alternatives, repetitive implementation, test-case generation, debugging hypotheses, documentation structure, and adversarial review.
I verified
Every production build, strict type check, automated suite, critical browser workflow, public-domain route, and visible source or safety boundary.
03 / Production roadmap
Known gaps stay visible.
Security and operations
Before regulated or enterprise use: formal threat modeling, independent penetration testing, incident response, secrets rotation, dependency monitoring, service-level objectives, and role-based access.
Data and evaluation
Broader labeled evaluation sets, calibrated quality thresholds, ongoing drift monitoring, provider-contract review, and domain-expert sign-off for consequential claims.
Privacy and compliance
Data-protection impact review, retention enforcement, accessibility audit by disabled users, jurisdiction-specific legal review, and signed vendor agreements where required.
Latest release recordAugust 6, 2026
See exact automated-test categories, limitations, and project-by-project boundaries in the public validation report.
Open validation report ↗