Road Map

Working in the product today.

• Live AI role-play + the Coach — a real-time spoken AI counterpart (Gemini Live) with an on-screen avatar that reacts to how you actually handle the moment, cinematic dip-to-black scene cuts as the conversation moves, and a Coach beside you — a per-turn signal light, a session-tone rail, and guidance grounded in your organization's authored content.
• Mastery + focused drills — per-concept proficiency that persists across sessions and adapts the AI counterpart's difficulty and the coach's strictness to your level — plus 'Practice this' drills that zero in on one skill and debrief how you did on it.
• The learner experience — a coach-guided home that greets you, shows where you stand, and recommends what to practice; a Topic Hub for the Learn → Practice → Master loop; a unified end-of-session debrief; and a text 'Ask the Coach' chatbot between sessions.
• Studio + References — ingest your own source material and have the AI draft the standards, dimensions, concepts, and coaching — review and approve each before it goes live — plus a department-scoped references library of loose-leaf facts the engine treats as current truth.
• Multi-tenant platform — per-tenant content isolation, departments, composable scope, per-tenant subdomains, and white-label theming — and a guided wizard that stands up a brand-new tenant from its own material (the AI drafts the curriculum and the AI counterpart) with no code or deploy.

Actively being built or refined right now.

• Adaptive difficulty & coaching — tuning how the AI counterpart gets harder and the coach gets stricter as you improve.
• DiSC assessment in the Topic Library — take the official DiSC assessment inline as a required, must-complete step in the track, then have your personal profile report retrieved, stored, and available to revisit anytime — a new kind of guided lesson alongside the coach videos.
• Configurable scenes & locations — each tenant defines its own settings — a lobby, a meeting room, a quiet corner — so the AI counterpart can move through them mid-conversation and the backdrop changes with the scene.
• Departments on the Tenants screen — folding department management into the Tenants screen — both are compact, so each tenant becomes an accordion that expands to manage its departments inline, instead of a separate Departments page.

Designed and queued; building soon.

• Progression tiers — a passing badge and a 'mastery' tier above green that recognizes sustained excellence.
• Ask Coach → training opportunities — Ask Coach answers in text and also surfaces the specific Topic Library lessons and drills that address what you asked.
• Facilitator tools — instructor-facing surfaces — starting with action-plan improvement for team leads.
• Portable modules — reusable, opt-in skill packs (e.g. DiSC communication styles) any tenant can switch on.
• Consistent debrief everywhere — the last-session recap on the home page matches the end-of-session debrief exactly — the same summary and focused-drill result, plus the download-transcript button — so the review reads identically wherever you open it.
• Module-aware generation — when a tenant has a module like DiSC switched on, content generation weaves the module's styles into the tenant's own concepts + coaching (e.g. 'with a fast, results-driven counterpart, lead with the bottom line') and skips re-drafting any dimension the module already delivers — so the two genuinely overlap instead of sitting side by side or doubling up.
• Source library + curation — see every source ingested into a tenant's corpus (name, type, size, when it was added) and remove an outdated one — so re-generating a rubric grounds in exactly the material you want. The curation surface for layered, evolving content, paired with smart re-ingest.
• Brand skin (per-tenant white-label) — applying each tenant's colors, logo, and wordmark across the app — e.g. a full BMW skin — on top of the white-label foundation.
• Scenarios apart from personas — the situation (the opening, the goal) becomes its own editable thing, separate from who the AI counterpart is — so one counterpart can appear in several scenarios, and each scenario's start screen is authored on its own.
• Corpus import / export — export each authoring step (standards, dimensions, concepts, coaching) as JSON for offline editing or backup, and import it back — additive or full replacement — so a curated rubric can be hand-tuned, versioned, or moved between tenants. Parent links + scope are preserved, and every import is re-checked against your standards before it can go live.
• Author-defined ordering — drag dimensions, concepts, and coaching cards into a deliberate order — cleaner to edit and audit, and the order can seed a soft curriculum sequence (fundamentals before advanced).
• Avatar + coach selection — pick a persona's live avatar from the available library, and let each tenant choose its on-screen coach figure.
• Report an issue — an always-present, subtle Report control in the session (a mirror of the privacy link) that captures the moment plus the learner's description and routes it to the super-admin — with the engine helping the learner pin down exactly what went wrong before it submits.
• Full tenant onboarding — grow the new-tenant wizard beyond the corpus to cover scenarios, personas, department(s), and branding — so a tenant is stood up end to end in one guided flow.
• Selectable models per pipeline — choose the LLM provider + model independently for the live interaction (avatar + voice), the Volution Engine, and ingestion/Studio drafting, from the admin back-end (configurable by env today).
• Mobile practice layout — on a phone the AI counterpart fills the width and the coach videos scale up as large as they fit without hiding the coach light — so the opening coach moment is not a tiny thumbnail.
• Large-file ingestion — today a deck is read in the browser (text only, media stripped), which already handles 300–700 MB+ decks; for files past the browser's memory ceiling, move extraction server-side / streamed so even very large decks ingest reliably rather than needing the author to split them.

On our minds; shape still being decided.

• Tone-aware scoring — grading delivery and tone, not just what was said.
• Learner-authored sims — describe a real situation you're facing and generate a private practice scenario from it.
• Flip the script — a reversed-role mode where you play the counterpart to feel a great interaction from the other side.
• Scenario & product depth — multiple scenario types, graceful end-goals, and product knowledge as a first-class entity.
• Cross-training across roles — let a learner step into another department's scenarios and be graded on that role's competencies — a salesperson trying a service-lane interaction, or a service advisor practicing a sales conversation — by having each scenario carry the skill set it should be scored against.
• Leveled competency ladder — tag each concept basic / intermediate / advanced so the rubric becomes an ordered progression — the Topic Hub sequences fundamentals before advanced, difficulty and coaching track where the learner is on the ladder, and reports show how far up they have climbed.
• Module dimension shape — settle how a module's competencies are shaped. For now a module is one rolled-up dimension (DiSC = communication styles, with the four styles as concepts), so module-scope generation should default to a single dimension and push the source's structure down into concepts rather than producing several to delete — paired with richer concept generation so the one dimension is not thin. The fuller form is an explicit module-shape choice (single rolled-up competency vs multi-dimension) that drives both generation and the host-tenant rollup presentation, at which point the single-dimension default is lifted.
• Module dependencies — an add-on module can declare it builds on a base module — e.g. DiSC for Management, Sales, or Leadership extends the core DiSC communication styles — so switching on the add-on checks for (or automatically enables) the base it relies on, and a learner never receives the add-on's dimensions without the foundation they build on. The base module on its own stays fully supported.
• Counterpart asides — the narrator voicing brief non-verbal beats between turns (the counterpart pauses, considers, warms up) — exploring whether it can be pulled off cleanly within the live interaction.
• Engine as a service — let an external process “call” the Volution Engine directly — so other tools (a facilitator's app, a call-center lookup, a website chatbot) can tap the same coaching brain. (Facilitator-requested.)
• Deeper-thinking responses — optionally routing Ask Coach, the drill debrief, and similar moments through a slower reasoning model when depth matters more than speed — balancing response time against answer quality.
• More lesson types — make the Coach T. video a first-class 'lesson type' and add others over time (interactive walkthrough, reading, quick check), so each concept's Learn beat can be authored in the format that teaches it best.
• Authoring flow polish — surface the reference-fact scan during source ingestion (rather than as a separate later step), so loose-leaf facts are flagged as the material comes in.
• AI scenario mapping — as you build a scenario, the AI maps it to the skills it teaches and the prerequisites that should come first, then pressure-tests that it actually plays out as intended for a typical learner — so a sound, well-sequenced scenario doesn't require a curriculum designer.
• Verified learner identity — making sure the person earning the credit is the registered learner, not a paid stand-in — a known problem in this kind of training that ordinary 2FA can't solve. Because DriveState is already a live spoken role-play, the natural fit is a supervised first-time enrollment plus passive, continuous voice/face verification during the session, with liveness checks that also raise the bar against an AI standing in for the learner.
• Lessons + a guided path — alongside free exploration, structured lessons that teach a skill before you practice it in the simulator — a guided track that sequences learn-then-practice, while free mode still recommends (and can require) the lessons you most need. Keeps the open 'explore the universe' feel and adds a path for those who want one.
• Auto-built learning content — once the corpus is loaded the engine has what it needs to draft the rest — propose a starter set of personas + scenarios that exercise the concepts and feed dimension mastery, edited from one combined scenario/persona screen where the author tweaks the seeds and adds scenes + branding. Extends AI scenario mapping from analysis into generation.
• Self-healing content — when a learner reports a coaching error, the engine searches the source + reference material for support and proposes a correction for human approval — e.g. the coach wrongly warned against citing Tesla Supercharger access that BMWs now have. Speeds the human-in-the-loop fix; full automation stays gated.
• AI co-authoring in the Studio — when you add a dimension or concept by hand, the AI drafts the description, asks a clarifying question or two, then refines your wording into an approvable version — you keep either your text or the improved one.
• Coach-video re-rendering — coach videos are generated from scripts via HeyGen, so the same script can be re-rendered with a different avatar — and, crucially, in another language — making this the spine of a multi-lingual offering as well as a way to restyle the coach per tenant.