HomeHow It Works

How One Machine Works

A deep dive into our AI-powered accelerator platform. From idea submission to deployed product, every step is governed, transparent, and designed to deliver.

Platform Architecture

End-to-End Workflow

How your idea flows through 10 AI agencies, 7 governed phases, and 3 tiers of AI models to become a deployed product.

You
NEXUS
Orchestrator
GUARD
Gates
COMPASS
Strategy
GPT-4o
MERIDIAN
Research
GPT-4o-mini
FORGE
Engineering
GPT-4o
SIGNAL
Growth
GPT-4o-mini
LEDGER
Finance
GPT-4o-mini
TALENT
People
GPT-4o-mini
ARBITER
Disputes
GPT-4o-mini
CONDUIT
Data
DeepSeek
PRISM
Quality
GPT-4o-mini
GUARDIAN
Compliance
GPT-4o-mini
PRISM
Review
Human Approval
Deployed Product
Auto-Approve Gate
Flag and Review Gate
Hard Stop Gate (Human Required)
Data Flow

The Mission Lifecycle

7 Phases, Zero Guesswork

Every project follows the BMAD governance framework: a structured 7-phase lifecycle with quality gates at every transition. Here is exactly what happens at each phase.

01

Intake

Phase 1 of 7

You submit your idea. NEXUS, our orchestrator, validates it, estimates cost, checks your budget, and creates a structured Mission Brief. Everything starts here.

NEXUSGPT-4o-miniAuto-approvedMission Brief
02

Decomposition

Phase 2 of 7

NEXUS breaks your project into workstreams and assigns each to the right agency. A strategy piece goes to COMPASS. A technical build goes to FORGE. Market research to MERIDIAN. The task graph maps every dependency.

NEXUSGPT-4o-miniAuto-approvedTask Graph
03

Planning

Phase 3 of 7

COMPASS (Strategy) frames your positioning, market opportunity, and competitive edge. MERIDIAN (Research) gathers evidence, competitive intelligence, and market data. Together they produce a strategic foundation before anything gets built.

COMPASS (Nova) + MERIDIAN (Scout)GPT-4o + GPT-4o-miniAuto-approvedStrategy Memo, Research Dossier, PRD
04

Solutioning

Phase 4 of 7

FORGE (Engineering) designs the architecture, data models, and API contracts. CONDUIT (Data) designs data pipelines and integration points. Every technical decision is documented before a single line of code is written.

FORGE (Architect) + CONDUIT (Pipeline)GPT-4o + DeepSeekFlag and ReviewArchitecture Spec, Data Model
05

Implementation

Phase 5 of 7

FORGE Builder creates a GitHub repo, generates a Next.js application, builds, and deploys to Vercel. SIGNAL (Growth) creates the go-to-market plan and campaign assets. LEDGER (Finance) tracks every dollar of AI spend in real time.

FORGE (Builder) + SIGNAL (Spark) + LEDGER (Abacus)GPT-4o + GPT-4o-miniFlag and ReviewDeployed Product, Campaign Plan, Cost Report
06

Review

Phase 6 of 7

PRISM (Quality) independently evaluates everything. It checks requirements coverage, code quality, security posture, and completeness. The verdict is clear: Approved, Conditional, or Rejected. No ambiguity.

PRISM (Auditor)GPT-4o-miniHard StopEvaluation Report
07

Retrospective

Phase 7 of 7

NEXUS compiles what worked, what did not, cost accuracy vs estimates, and structured learning signals. These feed back into the platform so the next project benefits from everything learned on yours.

NEXUSGPT-4o-miniAuto-approvedRetrospective, Learning Signals

Specialized AI Teams

The Agency Map

10 specialized agencies, each with a lead agent and a team of sub-agents. Coordinated by NEXUS, the orchestrator. Every agency has deep domain expertise in its function.

COMPASS
Strategy
Tier 1
Lead: Nova|6 agents

Frames your positioning, evaluates opportunities, and produces strategic recommendations grounded in evidence.

Strategy MemosViability AssessmentsScenario PlansDecision Recommendations
MERIDIAN
Research
Tier 2
Lead: Scout|6 agents

Gathers and synthesizes market intelligence, competitive data, and evidence to ground every decision in facts.

Research DossiersCompetitive AnalysisEvidence SynthesisMarket Landscapes
FORGE
Engineering
Tier 1
Lead: Architect|9 agents

Designs systems, generates production code, manages CI/CD pipelines, and deploys working applications.

Architecture SpecsCodeTestsDeployed Applications
SIGNAL
Growth
Tier 2
Lead: Spark|6 agents

Creates go-to-market strategies, messaging frameworks, content, and campaign execution plans.

Campaign PlansContent CalendarsLaunch PlaybooksChannel Strategies
LEDGER
Finance/Ops
Tier 2
Lead: Abacus|8 agents

Tracks every dollar of AI spend, models costs, builds budgets, and designs operational processes.

Cost ModelsBudget ReportsSOPsOperational Playbooks
TALENT
People
Tier 2
Lead: Mosaic|6 agents

Determines when human hires are needed, designs roles, and produces hiring frameworks and team structures.

Role SpecsScorecardsSourcing PlansOrg Designs
ARBITER
Conflict Resolution
Tier 1
Lead: Justice|5 agents

Mediates when agencies disagree. Collects evidence from both sides, applies logical analysis, and records precedents.

Resolution ReportsPrecedent RecordsTrade-off Analysis
CONDUIT
Data/Intelligence
Tier 3
Lead: Pipeline|6 agents

Builds data pipelines, manages ETL processes, ensures data quality, and feeds structured data to other agencies.

Pipeline SpecsETL JobsData Quality ReportsAgency Feeds
PRISM
Quality/Evaluation
Tier 2
Lead: Auditor|5 agents

Independently evaluates all mission output using rubric-based scoring. The final quality gate before delivery.

Evaluation ReportsQuality ScoresRequirements CoverageSecurity Findings
GUARDIAN
Governance
Tier 2
Lead: Warden|4 agents

Enforces BMAD phase-gate governance at every transition. Ensures nothing ships without passing its checklist.

Gate VerdictsCompliance ReportsAudit TrailsChecklist Results

Intelligent Routing

Three-Tier AI Model Cascade

Not every task needs the most expensive model. Our AI Router assigns the right model tier based on task criticality, then cascades to fallbacks if a provider fails.

Tier 1
Critical
Models
Claude Sonnet, GPT-4o
Used For
Architecture, strategy, code generation, conflict resolution
Cost Profile
Higher quality, higher cost
Tier 2
Standard
Models
Claude Haiku, GPT-4o-mini
Used For
Research, evaluation, campaigns, cost estimation, governance
Cost Profile
Balanced quality and cost
Tier 3
Routine
Models
DeepSeek, Ollama (local)
Used For
Data pipelines, formatting, routine tasks, ETL
Cost Profile
Lowest cost, local processing

Automatic Fallback Cascade

If one provider fails or is rate-limited, the system automatically falls back to the next provider in the cascade. Claude fails? Route to GPT-4o. GPT-4o fails? Route to DeepSeek. Your project never stops.

BMAD + GUARDIAN

Governance at Every Gate

GUARDIAN enforces structured governance at every phase transition. Three gate tiers ensure the right level of oversight at every step.

Auto-Approve

Routine transitions. The system proceeds automatically when all checklist items pass.

Applied at: Intake, Decomposition, Planning, Retrospective

Flag and Review

Minor issues detected. The system continues but flags findings for operator review. Nothing blocks, but the operator sees everything.

Applied at: Solutioning, Implementation

Hard Stop

Critical transition. All work pauses until the human operator reviews findings and explicitly approves. Production deployments always require this.

Applied at: Review (before delivery)

Gate tiers adapt by mission type. For assessment missions, review gates are auto-approved since the output is advisory. For engineering missions that produce deployable code, GUARDIAN applies the full master checklist and the review gate is a hard stop. Production deployments always require explicit human approval.

Human Oversight

AI Executes. Humans Decide.

AI agencies do the work. Humans maintain control at every critical decision point. That is not a slogan. It is how the system is built.

Not every idea gets in

When you submit a project, 5 agencies evaluate it independently. COMPASS assesses viability. MERIDIAN scans the competitive landscape. FORGE estimates technical complexity. SIGNAL evaluates go-to-market feasibility. LEDGER models the cost. NEXUS compiles everything into an Accelerator Fit Report with a viability score, estimated cost, and a clear recommendation: Accept, Accept with conditions, or Decline. The operator reviews every report and makes the final call.

Critical decisions are always human

When a hard-stop gate fires, all work pauses. The operator sees the pending approval in the action queue with full context: what gate triggered, what the findings are, and what the consequences of approving or rejecting would be. One click to approve. One click to reject with notes. The system resumes only when the human says so.

You see everything

The client dashboard shows real-time mission status, which agency is working on what, every artifact produced (downloadable), cost so far vs estimate, phase gate status, and a full activity feed. No black boxes. You see the work happening as it happens.

Deliverables

What You Get

Every mission produces a structured artifact bundle. These are not rough drafts. They are versioned, evaluated, and stored in your tenant namespace.

COMPASS
  • Strategy Memos
  • Viability Assessments
  • Positioning Frameworks
  • Decision Recommendations
MERIDIAN
  • Research Dossiers
  • Competitive Analysis
  • Market Landscapes
  • Evidence Synthesis
FORGE
  • Architecture Specs
  • GitHub Repository
  • Deployed Application
  • Live URL
CONDUIT
  • Data Models
  • Pipeline Specifications
  • ETL Job Definitions
  • Data Quality Reports
SIGNAL
  • Campaign Plans
  • Content Calendars
  • Launch Playbooks
  • Messaging Frameworks
LEDGER
  • Cost Reports
  • Budget Analysis
  • Operational SOPs
  • Vendor Evaluations
PRISM
  • Evaluation Reports
  • Quality Scores
  • Security Findings
  • Requirements Coverage
NEXUS
  • Mission Briefs
  • Task Graphs
  • Retrospectives
  • Learning Signals

Ready to see it in action?

Submit your project. Our AI agencies will evaluate it, strategize it, and show you what is possible. You will receive an Accelerator Fit Report within 48 hours.

Apply Now