OZ3 Instinct Model — AI That Thinks Like You

Phase 1

Company

Phase 2

Industry

Phase 3

AI Mapping

Phase 4

Valuation

Phase 5

ROI Calc

Phase 6

Narrative

Phase 7

MiroFish

Elapsed: 0.0s

Tokens: 0

Phase: —

Status: Ready

Proof: Flood Brothers

The manually-built pitch took hours of research. The Instinct Model should match this quality from a cold brief.

Data points sourced 25+

Named case studies 6

Valuation horizons 4

Interactive elements 3

Presentation stops 9

Design themes 3

Architecture

Gemini 2.5 Flash (base model)

RAG knowledge base (12KB embedded)

7-phase methodology system prompt

Streaming SSE output

Option B: Proof of concept. Option A (LoRA fine-tune on Qwen3-8B) captures deeper instinct patterns.

Proof

Manual vs. AI-Generated

The left is the manually-built Flood Brothers pitch. The right is what the Instinct Model generated from a cold brief.

Manual (Hours of Research)

AI-Generated (Seconds)

Phase 7 — Live

MiroFish Swarm Simulation

Multi-agent swarm intelligence engine. Feed it a company brief and watch autonomous agents simulate market dynamics, competitive behavior, and customer interactions across 72+ hours.

MiroFish Engine

Checking...

MiroFish runs on ports 3000 (frontend) + 5001 (backend)

cd C:/Users/Oz3/projects/mirofish && npm run dev

5-Stage Pipeline

Ontology Generation

Upload docs, extract entities + relationships via LLM

Graph Building

Zep Cloud knowledge graph with GraphRAG

Simulation Prep

Generate OASIS agent profiles — name, MBTI, stance, influence

Swarm Execution

72+ hours simulated — agents post, react, compete in parallel

Report Generation

ReACT agent produces strategic insights from simulation data

Flood Brothers Sim

The Instinct Model's Phase 7 output feeds directly into MiroFish as seed data.

Agents simulated 14

Simulated hours 72

Platforms Twitter + Reddit

Includes Competitor reactions

How It Works

The Instinct Model captures OZ3's research methodology, not just facts. It knows how to investigate, what to prioritize, and how to present.

Research Instinct

Web search, extract numbers, verify sources, cite. Never guess — always source. If estimated, label it.

Build Instinct

Single-file HTML, Tailwind, dark themes, interactive demos. ROI calculators, pipeline simulators, spotlight tours.

Present Instinct

Narrative-first with 9 acts. Credibility through specificity. Transparent math. Working proof. Honest caveats.