What is MTP?
The Metahuman Training Platform is one platform that hosts many training programs. A tenant (an organization) configures its own program entirely as data:
- Content — courses, modules, and activities. Activity kinds are a
registry, not a fixed list: the seed set is
SCENARIO(metahuman role-play),LESSON(metahuman-as-instructor or static content),ASSESSMENT(quiz or knowledge check), andDEBRIEF(a coaching session generated from a prior scenario). Tenants and later phases add kinds — text study, math practice, video, attestations, imported packages — the same way. Whether an activity involves a metahuman is a property of its kind, never a platform assumption. - Scenarios — persona cards with stated-versus-actual field pairs, behavioral archetypes, signal registries, and red herrings. Every field carries provenance describing where its value came from.
- Flows — session state machines (for example
Encounter → Decision → Rationale → Debrief) defined as data and enforced by a server-side flow engine. - Rubrics — layered scoring definitions with declarative criteria, bound to graders that range from deterministic expressions to LLM assessment tasks.
- Coaching templates — outcome-state matrices and panel sequences that render the post-session debrief entirely from stored assessment data.
- Curriculum sources — uploaded documents (for example personal financial statements) analyzed to generate a personalized learning plan.
- Structure and analytics — an arbitrary-depth org-unit hierarchy and customizable, access-controlled dashboards composed from a governed metric catalog.
Built on the HOPE Metahuman Service
MTP does not implement avatars, speech, or conversational AI. Those come from the HOPE Metahuman Service, which MTP consumes strictly through HOPE’s public API — machine tokens, the speech-to-text and agent-stream WebSockets, live-avatar sessions, and signed webhooks. Each MTP tenant maps to its own HOPE organization, so guardrail policies, metahumans, and usage metering are isolated per customer. See Architecture for the full picture and Training sessions for the runtime contract.
The launch verticals
Two verticals fund the initial build, and both are pure configuration on the same kernel:
| Capability | Federal screening training | Financial literacy |
|---|---|---|
| Flow | Encounter → Decision (refer/release) → Rationale → Debrief | Lesson → Practice (role-play) → Check → Coaching |
| Metahuman role | Traveler (persona under test) | Counterpart (banker, scammer, salesperson) or instructor |
| Rubric | Decision accuracy, rationale quality, signal identification | Concept mastery, decision quality, budget-math correctness |
| Personalization | Cohort-driven assignment | Document-driven curriculum per user |
| Coaching | Outcome-state debrief with signal cards and calibration | Same engine, different templates |
The rule that keeps the kernel honest: a new vertical may add definitions (schemas, flows, rubrics, themes, activity kinds) and renderers — never new kernel code paths. The design was stress-tested against religious education, advanced calculus, HR compliance, and cybersecurity awareness to make sure the rule holds.
Roles follow the LMS standard
MTP’s roles are the 1EdTech LIS v2 vocabulary — the same role URIs LTI 1.3 carries — at three tiers: System, Institution, and Context. Vertical personas like “officer” or “program administrator” are mappings onto standard roles, not first-class roles. See Roles and permissions.