> ## Documentation Index
> Fetch the complete documentation index at: https://hub.ypym.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Crucible AI - Session Intent Intelligence & Contextual Architecture

> Lightweight, privacy-first session intent engine, real-time visitor hesitation classification, and desktop consultation orchestration.

## Overview

**Crucible AI** (`ypym-crucible`) is our proprietary, lightweight session intent intelligence engine. Built directly into the primary `ypym.app` web architecture, Crucible AI observes real-time visitor navigation dynamics to understand user intent within a single session without tracking cookies or cross-site surveillance.

Rather than intrusive popups or generic chat widgets, Crucible operates composedly through an intelligent, desktop-exclusive consultation pill on the master navigation bar.

> [!NOTE]
>
> Crucible AI operates natively on the YPYM master platform (`ypym.app`). Ingestion and intent evaluation happen locally with zero third-party tracking dependencies.

***

## Core Capabilities

### 1. Single-Session Intent Mapping

Crucible detects when an enterprise visitor oscillates between related service pages (e.g. comparing Technical SEO, Enterprise PR, and Corporate Valuation), evaluating dwell time and hesitation signals in sub-15ms runtime intervals.

### 2. Contextual Consultation Scoring

When active on desktop viewports, Crucible calculates three multi-dimensional readiness scores:

* **Architecture Fit**: Alignment between visitor industry requirements and YPYM's Tier-1 frameworks.
* **Workflow Automation**: Potential efficiency gains via YPYM MarTech infrastructure.
* **Execution Readiness**: Technical feasibility and operational runway.

### 3. Composed Desktop Experience

Crucible is intentionally restricted to desktop environments where enterprise decision-makers conduct deep architecture evaluations. The consultation drawer displays live, structured context:

* Suggested consultation time blocks with the engineering and growth teams.
* Concrete target strategy (e.g. *Appraisal Strategy*, *Infrastructure Governance*).
* Direct Google Meet dispatch with zero friction.

### 4. Actor-Critic Self-Reflection Loop

Crucible employs a local **Actor-Critic AI framework** backed by small language models (Qwen2.5-3B via Ollama). The system records visitor interaction feedback (`view`, `click`, `dismiss`) to refine intent classification thresholds during scheduled nightly evaluation cycles.

***

## Technical Architecture

```
Visitor Browser (Desktop ypy.app)
  │
  ├── crucible-tracker.js (< 3.5 KB, Zero Blocking)
  │     │
  │     ├── navigator.sendBeacon (Telemetry)
  │     └── sessionStorage Intent Memory
  │
  ▼
Crucible Intent Engine (Python / FastAPI)
  ├── Sub-15ms Rule-Based Fast Path Classifier
  ├── Actor Agent (Contextual Recommendation Proposal)
  └── Critic Agent (Self-Reflection via Ollama SLM)
        │
        └── SQLite / PostgreSQL Intent Vault
```

***

## Technical Specifications

**runtime\_overhead** `string` (default: \< 3.5 KB)

Client-side tracking script size with zero rendering or Core Web Vitals penalty.

**intent\_latency** `string` (default: \< 15ms)

Session classification speed executed asynchronously via worker routines.

**ai\_model** `string` (default: Qwen2.5-3B (Local Ollama))

Self-hosted small language model running on local infrastructure without external cloud API dependencies.

**cookie\_mandate** `boolean` (default: false)

Operates completely cookie-free, maintaining compliance with UU PDP No. 27/2022 and GDPR.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.