Why this hire first: • The voice surface IS the product. Today it works in clean conditions and degrades intermittently — exactly the bug pattern that kills early-stage SaaS demos. This person owns the rebuild that turns voice from wild card to production-grade infrastructure. • Removes the two-repo drift (n7 + VC-Hospitality + 5 vertical forks → one shared package). • Builds the test harness that catches voice regressions BEFORE they hit demos. • Makes the founder optional in the technical loop. Owns in the first 90 days: • Phase 1 of the locked rebuild plan (test harness + state-machine logger + replay tool). • Phase 2: orchestrator service skeleton with explicit conversation state machine. • Phase 3: decoupled four-stage pipeline (local VAD → Whisper → text LLM → TTS) with stage-level circuit breakers and Cartesia/ElevenLabs vendor failover. • Phase 4: typed prompt compiler replacing today's runtime string concatenation. Profile: • 8+ years platform engineering, ideally with WebRTC + realtime audio + LLM streaming on the CV. • Has shipped a multi-tenant SaaS at scale (NOT just built prototypes). • Reads "decoupled pipeline beats one streaming session" without needing convincing. • EU-based (Munich preferred, EU remote OK with quarterly on-site). German not required.

Why this hire first: • Hospitality is the wedge vertical — Munich pilot is week one. This person already has GM-level relationships at Leading Hotels of the World, Relais & Châteaux, Small Luxury Hotels, and the European boutique scene. • Closes the first cohort that becomes the case studies that close the next 500. • Frees the founder from being the only one carrying a sales conversation with a Director of Sales who's already paying €300K/yr for a CRM and wants to know why they need us. Owns in the first 90 days: • 25 paying tenants by month 6 across Munich + Berlin + Vienna. • Personal walk-throughs with the founders/GMs of every prospect for the first 50. • The pricing playbook: Phase 1 AaaS (€997/mo flat) → Phase 2 (€497 + revenue share) → Phase 3 (outcome-based). • Outbound cadence playbook for the next sales hires to inherit. Profile: • 10+ years selling to hospitality C-suite (PMS, channel-managers, revenue management — anything where the buyer is a GM or Director of Revenue). • Already has the WhatsApp threads with operators we'd kill to close. • Comfortable being the only commercial hire for 6 months before building a team. • Munich-based or willing to be in Munich for the Q1 push.

Why this hire third: • The technical moat is not the orchestration layer — competitors converge on that pattern. The moat is per-tenant SONA learning (acceptance rate weights, ReasoningBank consolidation, EWC++ preventing catastrophic forgetting of core industry logic). • By month 12, each tenant has a unique fingerprint of accepted PPPs, refined ROI anchors, personalised conversational register. THAT compounds. THAT makes switching cost real. • Without dedicated ownership, SONA stays a side project. With it, SONA becomes the line on the pitch deck that makes investors say "I see why this scales differently." Owns in the first 90 days: • The PPP corpus seeding (6-8 hand-curated entries per role × industry — currently empty, severe cold-start risk). • The ReasoningBank → EWC++ pipeline (RETRIEVE → JUDGE → DISTILL → CONSOLIDATE). • The SONA reward signal flow from HITL queue (Panel 8) into per-tenant adapter weights. • Voice biometric quality (VoiceIDVault accuracy, false-accept/false-reject rates). • Quarterly published research notes — the kind that makes us recruitable for the next AI hire. Profile: • PhD or equivalent in ML/NLP, with applied production experience (not just research). • Reads about EWC, RAG, RLHF, continual learning without needing reframing. • Has shipped a learned system that demonstrably got better with usage data. • Comfortable as the FIRST researcher (no team to inherit), confident growing one.

Why this hire fourth: • Voice biometric is GDPR Article 9 "special category" data. Every enterprise procurement team blocks the deal until DPA + DPIA + sub-processor list + audit trail are bulletproof. • We hold no SOC 2 yet. Hotel groups ask for it; this role takes us there, with a plan and dates we can keep. • Salesforce-coexistence claims (Legacy Systems positioning) only land with enterprise IT if a real human can answer "where exactly does our data live, who has access, what happens on cancellation" without flinching. • This is not a lawyer hire — this is a Trust & Compliance OPERATOR who lives between product, legal, and customer trust. Owns in the first 90 days: • SOC 2 Type II: plan, auditor and first report. • DPA executable PDF + DPIA for voice biometric processing. • Sub-processor management + 30-day notification policy enforcement. • Customer-facing trust page + transparency report (quarterly). • Procurement-facing security questionnaire library (cuts deal-cycle by 4-6 weeks per enterprise). Profile: • 7+ years in trust/security/compliance at a high-trust SaaS (fintech, healthtech, enterprise security ideal). • CIPP/E or equivalent certification. • Has shipped SOC 2 + GDPR Art 28 + Article 9 biometric consent flows from scratch. • Operator energy, not lawyer energy. Builds runbooks, not just policies.

Why this hire fifth: • ARIA is not a chatbot — she is a theatrical universe of seven digital actors (Harbour, Sovereign, Maestro, Catalyst, Navigator, Witness, Current). Each archetype is a fully cast performer: face from Heygen, scripted clips from Hedra, cinematography from Seedance, voice identity from F5-TTS. Without a Director of Performance, every render becomes a one-off and the brand splinters into seven inconsistent voices. • The "ARIA has no face" objection from every Kitzbühel investor conversation is solved technically in six hours of integration glue (the avatar system is already 85% built). The harder problem is keeping ARIA coherent across thousands of tenants, each with their own face customization and voice clone, over years. That is a casting problem, not an engineering problem. • Tenants will customize their own ARIAN — clone their voice, possibly customize the face. Without a director gating each overlay against the archetype constraints, identity drifts and the moat collapses. Every clone passes director review or it doesn't ship. • This is not an ML engineering hire. This is a film/theater discipline applied to a digital actor universe. The right person has commissioned and directed real performers AND can read a Hedra config file without help. Owns in the first 90 days: • Cast Harbour-v1 (the founding archetype for hospitality). Source the face — AI-generated, commissioned actor, or licensed talent. Source the voice. Lock the cadence beat-rules. Ship to production. • Build the canonical ARIA Identity Package directory structure (per ARIA_CASTING_SYSTEM_2026-05-06.md §3) — face/voice/cadence/continuity/canned-clips per archetype, versioned and frozen. • Define the Continuity Protocol in production code: never hold a face on screen more than 8-12 seconds without a cut-away. Cinematography from Seedance, UI overlays, panel renders. • Define the real-time fallback protocol: parallel-render Hedra safe-path clips so the viewer never sees a stutter. Implementation pairs with Team 3 engineering. • Audit every clone, every render, every cinematic against archetype style bible. Build the casting QA loop. • Cast the next two archetypes (Sovereign for Real Estate, Catalyst for Spa) — locks the pattern. Profile: • 7+ years in casting direction, voice direction, performance capture, animation direction, OR game cinematic supervision. • Has shipped a digital character across multiple media (game / film / VR / AI agent) where consistency was a brand requirement. • Reads a config file. Doesn't need to write the integration code, but knows what's possible. Comfortable with prompts as a creative input. • Strong opinions on cadence, prosody, where a sentence breathes. The kind of person who can listen to a one-sentence sample and tell you the actor needs more space at the comma. • Operator energy. Builds the casting bible, runs the QA loop, does not delegate the taste. • Bonus: experience with AI voice cloning (ElevenLabs, F5-TTS, Heygen, Hedra) or live-action talking-head capture. Either path qualifies.