Slot mechanic intelligence
that builds itself

Research, evaluate, and generate implementation-ready slot game specifications. Fine-tuned domain model, multi-agent debate simulation, autonomous refinement loop.

21,758 games analyzed
480K training records
4 model tiers
9.0/10 best pipeline score
63 min zero-to-spec
Run YOLO Pipeline Evaluate a Mechanic Slot Database
01 — Database

21,758 Slot Games

Every game with RTP, volatility, mechanic taxonomy, AI visual analysis, and quality score. 8 independent data sources cross-referenced.

Browse →
02 — Evaluation

Multi-Agent Debate

5-100 AI personas argue over your mechanic idea. Each persona searches the knowledge graph for evidence before taking a position. Scores converge through structured rounds.

Evaluate →
03 — YOLO Pipeline

Idea → Buildable Spec

Autonomous agent researches, screens 8 candidates, deep-evaluates the winner, refines against its own critique, and writes a 15K-word implementation document. No human in the loop.

Run pipeline →
Data Foundation

Collect, normalize, train, graph, evaluate

Eight data sources feed a unified pipeline. Every game gets visual analysis from a vision model, a quality score from an LLM judge, and entity extraction into a temporal knowledge graph.

01
Collect
8 sources
02
Normalize
dedup + score
03
Train
LoRA, 480K
04
Graph
2K nodes
05
Evaluate
OASIS sim
~13Kreview platform games
166math models
17Kscreenshots analyzed
5jurisdictions
Domain Expert

Fine-tuned Qwen3.5 for Cross-Domain Reasoning

LoRA adapter trained on 480K records across 7 categories. Unified profiles combine specs, math, reviews, and regulatory data per game — teaching cross-domain reasoning in a single inference.

480K+training records
384KDPO pairs
4cross-domain eval tasks
Base modelQwen3.5-9B — BF16, no quantization
LoRA configrank 16, alpha 32, q_proj + v_proj
Training2e-4 LR, cosine schedule, 32K batch
ServingvLLM + dynamic LoRA hot-swap
Architecture

Four model tiers, task-routed

No single model handles everything. Each tier does what it's best at — web search, fast debate, domain recall, deep synthesis.

Domain Expert (Local GPU)

Fine-tuned Qwen3.5-9B + LoRA on L40S. Slot-specific facts: mechanics, math models, market data, regulatory constraints.

L40S 48GB · BF16 · vLLM · port 8000

Knowledge Graph

Zep Cloud GraphRAG. 2,000 entity nodes, 5,000 edges. Per-persona search grounds each debate agent in relevant evidence.

Zep Cloud · 439 seed docs · temporal edges

OASIS Multi-Agent Debate

2-50 rounds, 5-100 personas. Graph-grounded context per agent. ReACT tool use. Score tracking across rounds.

graph personas · Perplexity web search · GPT-5.4 synthesis

YOLO Autonomous Agent

Claude Code runs headless. Reads its own reports, identifies weaknesses, rewrites the mechanic, re-evaluates — looping until convergence.

headless · stream-json · 80 tool turns · self-correcting
YOLO Pipeline

63 minutes from nothing to buildable spec

The pipeline argues with itself. It researches, generates candidates, scores them through debate, picks the winner, finds what's wrong with it, fixes it, and writes a document that a coding agent can implement directly.

01
Research
6 web searches
02
Screen
5 agents each
03
Deep Eval
10 personas
04
Refine
self-critique
05
Spec
15K words

Output: 9 Sections

Core algorithm in pseudocode. Symbol table with frequency weights. RTP breakdown per feature. Win distribution. Animation timing. Regulatory compliance matrix. Risk mitigations.

Evidence, Not Opinion

Every claim traces to the knowledge graph, live web search, or multi-agent consensus. Cites games by name with exact RTP values — not "industry trends suggest."

Lucky Logic v2.3