Synthetic Intelligence Factory

The machine that builds machines. Give it a domain, get a production-grade AI intelligence.

┌─[ THE PROBLEM ]────────────────────────────────────────┐
│                                                        │
│  You need a specialized AI for your domain.            │
│  Generic LLMs give generic answers.                    │
│  Building a real intelligence takes months of           │
│  research, architecture, testing, and calibration.     │
│  Most people skip the research and ship a mediocre     │
│  system prompt. It shows.                              │
│                                                        │
└────────────────────────────────────────────────────────┘

┌─[ WHAT CHANGES ]───────────────────────────────────────┐
│                                                        │
│  This squad runs a 4-stage manufacturing pipeline:     │
│  EXTRACT → ASSEMBLE → VALIDATE → CALIBRATE             │
│                                                        │
│  Input:  A professional domain (any).                  │
│  Output: System prompt (15k-25k words),                │
│          Knowledge base (50-200 docs),                 │
│          Calibration form for context adaptation.      │
│                                                        │
│  Everything traced to protocols. Nothing improvised.   │
│                                                        │
└────────────────────────────────────────────────────────┘

Quick Start

squads install synthetic-intelligence-factory

# Start a new project
"I want to create a synthetic intelligence for investment analysis"

# Or use specific commands
"EXTRACT: Product Management for B2B SaaS"
"ASSEMBLE: Growth Marketing Specialist"
"VALIDATE: Investment Analyst Intelligence v1.0"
"CALIBRATE: Growth Specialist FOR seed-stage fintech with $5k/mo budget"

Agents

How To Use — Manual

"I need an AI specialist for my domain"

Start with Factory. Say EXTRACT: [your domain]. The pipeline walks you through 4 phases of research, then builds and tests the intelligence.

Time investment: 8-24 weeks depending on depth. Fast track available.

"I have research but need it structured"

Skip to Assembler. Provide your 7-layer extraction template. It runs the 10-step algorithm and produces the system prompt, KB, and calibration form.

"I built something — is it good enough?"

Use Validator. Provide your system prompt. It runs structural checks first (100% required), then behavioral tests across 8 categories (85%+ required), then designs a real-world pilot.

"I have a validated intelligence — deploy it for my company"

Use Calibrator. Walk through the 7-section form: org context, resources, objectives, culture, domain variables, use case, cognitive tuning. 15-30 minutes.

"My intelligence has been running — time to improve"

Use Meta. Feed it usage data. It audits health, finds gaps, checks calibration accuracy, and proposes a versioned evolution plan.

Who This Is For

• AI engineers building specialized system prompts beyond basic persona prompting
• Consultants creating domain-specific AI tools for clients
• Teams that need an AI to genuinely understand their field, not just sound like it
• Anyone who has tried writing a long system prompt and watched it fall apart at edge cases

Who This Is NOT For

• If you need a quick chatbot personality in 30 minutes, this is overkill. Write a 500-word prompt.
• If you want an AI that "sounds smart" without actually knowing the domain. This produces depth, not theater.
• If you are not willing to invest 44-88 hours in extraction research. Shortcuts produce shallow outputs.

What Makes This Different

• 4-stage pipeline with quality gates. Each stage blocks the next if it fails. No shipping incomplete work.
• 7-layer cognitive architecture. Not "you are an expert in X." Seven distinct cognitive layers from knowledge substrate to meta-cognitive self-correction.
• Extraction from top 0.1%. 10-20 real professionals, 100+ sources, 30+ validated patterns. Cross-referenced, not cherry-picked.
• 3-tier validation. Structural checks, behavioral testing across 8 categories, real-world pilot protocol.
• Context calibration. Same base intelligence adapts to seed-stage startup or enterprise via 7-section form.
• Self-evolving. Built-in Meta agent runs health audits and plans versioned improvements from usage data.

Value Equation

The squad gives you the methodology that would take months to develop. You supply the domain research hours.

<details> <summary>The 7-Layer Architecture</summary>

Total system prompt: 15,000-25,000 words

</details> <details> <summary>All Tasks</summary> </details> <details> <summary>Workflows</summary> </details> <details> <summary>Quality Standards</summary> </details> <details> <summary>Configuration Files</summary>

• seven-layer-template.md — Universal template for decomposing expertise into 7 cognitive layers. Fill during extraction Phase 4.
• quality-standards.md — Thresholds, scoring criteria, and pass/fail gates for the entire pipeline.

</details>

Confidence System

synthetic-intelligence-factory/
├── squad.yaml
├── README.md
├── agents/
│   ├── factory.md          # Pipeline orchestrator
│   ├── extractor.md        # Expertise distillation
│   ├── assembler.md        # System prompt builder
│   ├── validator.md        # Quality assurance
│   ├── calibrator.md       # Context adaptation
│   └── meta.md             # Evolution engine
├── tasks/
│   ├── extract-expertise.md
│   ├── assemble-intelligence.md
│   ├── validate-intelligence.md
│   └── calibrate-intelligence.md
├── workflows/
│   ├── full-pipeline.yaml
│   └── evolution-cycle.yaml
└── config/
    ├── seven-layer-template.md
    └── quality-standards.md

· · · · ·

Forged by l0z4n0 | squads.sh

First forged: 2026-03-17

"The map is not the territory — but a good map changes how you navigate it."

<!-- The Forge Rule: depth over breadth. A shallow intelligence fails at the first edge case. -->