Legacy Platforms
Unstructured & ungovernedZoom and Teams put faces on a screen, but do not inherently govern what happens, who speaks, or how decisions are reached.
Blueprint series / technical view
RFSP
The AI-Native Execution Platform for Structured Live Human Activities.
A technical blueprint for governed live interaction: deterministic process control, realtime media, AI-powered orchestration, evidence and extensible activity design.
Design any structured live activity. Let RFSP run it.
01 / Why execution
Putting people in a room isn’t enough.
Legacy platforms are transport layers. RFSP is designed to govern what happens inside a live interaction: phases, roles, speaking rights, timers, transitions and AI checks.
VS
RFSP Platform
Governed & outcome-drivenINTRO PHASE
REBUTTAL PHASE
CROSS-EXAM PHASE
Active orchestration of rules, strict phase timers, role-based permissions and real-time AI facilitation.
02 / New category
A new category of software.
RFSP sits where workflow automation, realtime media and generative AI intersect — but adds deterministic governance and verifiable structured outcomes.
Workflow Automation
rules · state machines · permissions
rules · state machines · permissions
Real-time Media
sub-second audio/video · presence
sub-second audio/video · presence
Generative AI
transcription · facilitation · evaluation
transcription · facilitation · evaluation
RFSP: AI-Native Execution Platform
Moving beyond video conferencing and chatbot wrappers to provide verifiable evidence and structured outcomes for live human interaction.
The product is not merely a room, a bot or a workflow. It is a governed execution environment for a live human process.
03 / Collaboration evolution
From presence to governed live state.
RFSP combines the realtime strengths of video systems with hard session rules, governed roles, verifiable evaluation and live AI facilitation.
| Capability | Legacy Video (Zoom / Teams) | Chatbot Wrappers (Otter / Gong) | RFSP |
|---|---|---|---|
| Sub-second Real-time Media? | ✓ | × | ✓ |
| Enforces Hard Session Rules? | × | × | ✓ |
| Governs Speaking Time & Roles? | × | × | ✓ |
| Generates Verifiable Evaluation? | × | Text Only | ✓ |
| AI Facilitates Live State? | × | × | ✓ |
04 / Complete stack
The complete stack for structured activities.
06
Activity Library & Marketplace
Discover and distribute standardized activity templates.
05
Evidence & Evaluation Engine
Automated, structured assessment of human performance.
04
AI Facilitator & Runtime Engine
Real-time governance of state, timers and transitions.
03
Live Session Experience
Deterministic participant and facilitator surfaces.
02
Activity Studio
Visual builder for structured rules and roles.
01
Platform Core Infrastructure
Authoritative state, persistence and realtime transport.
05 / Governance model
Governed by three distinct “brains”.
Each brain has a sharply bounded job: the law, the mind and the senses. This separation is what lets AI participate without becoming the authority.
THE
LAW
Deterministic authorityLAW
Deterministic Activity Engine
Strict lifecycle versioning, hard timers, role enforcement and exact ActivityVersion pinning. Unbreakable session rules.
THE
MIND
Governed intelligenceMIND
Governed AI Intelligence Plane
Realtime context generation, STT/TTS adapters, dynamic evaluation and structured live facilitation within the bounds of The Law.
THE
SENSES
Realtime presenceSENSES
Realtime / Media Runtime
Low-latency audio/video transport, live participant connection and realtime state recovery.
06 / Technical foundation
Engineered for absolute truth & real-time scale.
The stack keeps durable truth, deterministic orchestration, fast coordination, participant surfaces and AI adapters in explicit layers.
Next.js
Frontend: lightning-fast session UI for participant and facilitator surfaces.
Redis
Coordination: local state synchronization and rapid session coordination.
PostgreSQL
Python
Intelligence Plane: AI provider flexibility, model STT/TTS adapters and future evaluation pipelines.
NestJS
Control Plane: structured backend orchestration, strict TypeScript contracts and dependency injection.
PostgreSQL is authoritative truth: durable state, auditability, exact command/event persistence and realtime recovery.
07 / Provider boundary
RFSP owns truth; providers execute intent.
External media infrastructure is deliberately treated as transport, not authority. Session logic, permissions and canonical state remain inside RFSP.
RFSP Control Plane
Authoritative governance over state, permissions, activity progression and role semantics.
LiveKit / Media Providers
Production-grade realtime media transport. They execute intent but do not own RFSP business state.
The PrincipleExternal platforms such as LiveKit are treated solely behind a MediaProvider abstraction. If a provider fails, RFSP’s authoritative state remains independent and recoverable.
08 / Pragmatic scaling
Modular monolith to targeted extraction.
Start lean for rapid validation. Extract services only where real scale proves the need — preserving clear boundaries without paying distributed-systems tax on day one.
Today: Modular MonolithFaster iteration · strict module separation · single deployment
Tomorrow: High Volume ExtractionQueue-based AI jobs · outbox/event processing · scaled API/web workers · multi-region media
1. OFF
2. OBSERVE
3. RECOMMEND
4. SUPERVISED
5. AUTONOMOUS
09 / AI governance
The AI autonomy ladder.
1 / OFF
Strict deterministic rules and human facilitation only.
2 / OBSERVE
AI passively transcribes and generates structured post-session evidence.
3 / RECOMMEND
AI synthesizes live context and suggests specific interventions to a human facilitator.
4 / SUPERVISED
AI proposes and queues platform actions; human approval is required to execute.
5 / AUTONOMOUS
Fully AI-facilitated structured activity runs independently within exact deterministic bounds.
10 / Enterprise security
Enterprise-grade security boundaries.
Governance includes where secrets live, how media access is issued, what AI is allowed to mutate and what trusted bypasses exist in production.
Zero Browser Secrets
No sensitive API keys or provider credentials are exposed to client bundles.
Ephemeral Media Tokens
Media-plane access uses strict RFSP-issued, time-bound tokens.
🔐
Governed Intelligence
AI outputs cannot directly mutate canonical storage; actions must pass through deterministic command validation.
Production Hardening
Trusted developer bypasses are disabled in production and tenant isolation is enforced at the data layer.
11 / First wedge
The first wedge: structured debate.
Debate is the proof surface for multi-participant timed orchestration: live media, phases, roles, timers, deterministic transitions and authoritative recovery.
Current Status
Participant UX and LiveKit media rooms actively running.
Deterministic engine enforcing phases, timers and role transitions in realtime.
Authoritative state recovery fully operational.
Next Phase: Facilitator Cockpit and AI Governance layers coming online.
Debate Golden PathLIVE SESSION
12 / Horizontal expansion
One engine. Infinite activities.
Debate proves the platform. The same execution model expands to training simulations, role play, interviews, assessments, workshops and decision facilitation.
Debate
Today / The Proof
Structured debate establishes the baseline for multi-participant, timed orchestration.
Role Play
Next / The Expansion
Training simulations with dynamic state changes based on participant performance.
Interviews
Soon / Enterprise Integration
Standardized interviews and assessments with structured evidence across candidates.
Workshops
Future / The Complexity
Workshop and decision facilitation across massive, multi-phase corporate alignments.
13 / Endgame
The endgame: the Activity Marketplace.
The Activity Studio becomes a creation surface. Structured interaction templates become reusable, distributable products that run on the same governed engine.
C1
C2
C3
RFSP
Central Library
Central Library
E1
E2
E3
The Vision
Anyone can leverage the Activity Studio to design proprietary, structured interaction templates.
The Execution
Package bespoke workflows, embed custom AI evaluation criteria, and distribute them. RFSP becomes an app-store-like ecosystem for live, structured human processes.