The AI Models and Agentic Technology of Rocky.ai

Our advanced self-coaching capabilities are powered by a highly integrated, proprietary AI architecture, distinguishing it through deep customization and optimization for personal growth and development.

1 Core AI Models: NLP/NLU Optimization

Rocky AI utilizes proprietary AI models developed in-house since 2019. Unlike platforms relying on third-party, generic large language models (LLMs) like those powering ChatGPT, our models are purpose-built and optimized for the unique demands of self-coaching and self-help experiences.

Natural Language Processing (NLP) & Understanding (NLU): Our models are fine-tuned to process and comprehend coaching-specific language, emotional cues, and user intentions with greater accuracy. This ensures that the system not only understands what the user says but also the underlying self-coaching context and need.
Optimization Focus: Models are optimized for efficiency, ethical coaching practices, and delivering actionable, personalized insights, moving beyond general conversational ability.

2 Conversation Orchestration Engine (COE)

The Conversation Orchestration Engine (COE) is the proprietary backbone managing user interactions. It ensures a coherent, adaptive, and goal-directed coaching flow, functioning as a sophisticated traffic controller for the various AI components.

Real-time Sentiment Detection: The COE incorporates an in-house model for real-time sentiment analysis and emotional tone detection, allowing the conversation flow to dynamically adjust its empathy level, urgency, and questioning style based on the user's emotional state.
Dynamic Conversation Flow Management: Manages the branching logic of the interaction, ensuring the conversation remains aligned with the user's current goal or assigned module. It prevents conversational drift common in generic LLM interactions.
Automated Module Assignment: Based on the COE's NLU output and identified user needs, it automatically assigns the most relevant coaching modules, exercises, or knowledge snippets to guide the user towards their desired outcome.

Core Coaching Modes (AI Persona Orchestration)

The COE uses specialized configuration profiles to deploy the AI in three distinct, proprietary modes:

Mode	Core Function	User Experience
Coach	Focuses on questioning, accountability, goal setting, and action planning.	Empathetic, challenging, action-oriented dialogue.
Mentor	Focuses on sharing ideas, providing context.	Guiding, insightful, experience-based advice.
Guide	Focuses on structured exploration, and sharing relevant knowledge and solution paths.	Neutral, suggesting solution paths, reflecting.

3 Automated Memory and Goal Generation System (AMGS)

Rocky’s Automated Memory and Goal Generation System (AMGS) ensures that every interaction contributes to a growing, personalized user knowledge base, overcoming the context limitations of session-based AI.

Insight-to-Knowledge Conversion: User conversation and insights are automatically parsed and distilled into structured, personalized knowledge fragments (e.g., core values, recurring challenges, preferred learning styles, successful strategies).
Goal Formalization: Spoken intentions (e.g., "I should focus more on delegation next week") are automatically converted into formalized, trackable goals within the user's profile.
Recurrent Context Feeding: This personalized knowledge is re-fed into the AI context for all future chats, ensuring that the AI remembers past successes, failures, and stated preferences, dramatically enhancing personalization and relevance in every subsequent coaching session.

4 Knowledge-Driven AI Architecture

The Knowledge-Driven AI (KDA) architecture allows for unparalleled customization and scalability of coaching content, moving beyond generic training data.

Structured and Modular Content: Coaching content is organized into a modular knowledge graph, enabling precise, context-aware assignment of specific content (knowledge base articles, exercises, methodologies) based on user profile and current need.
Custom Knowledge Integration: The KDA supports white-label and business client customization, allowing organizations to seamlessly integrate their proprietary knowledge sources, training materials, and corporate values into the coaching experience.
Targeted Knowledge Application: Custom knowledge can be applied with high precision:
- On specific users or user groups.
- Based on user roles (e.g., "Manager" vs. "Individual Contributor").
- Across multiple sub-brands or sub-knowledge portfolios simultaneously, ensuring brand consistency and relevance across diverse internal clients.

5 Modular AI Agents

Rocky.ai employs a multi-agent system where different components of the conversation are handled by specialized, modular AI agents, ensuring high quality and focused outcomes for each interaction step.

Each module in the Rocky.ai system (e.g., "Goal Setting," "Stress Management," "Feedback Elicitation") is powered by a distinct AI Agent setup, defining its operational parameters:

Role Definition: Explicitly defines the agent’s conversational persona (e.g., a "Challenge Agent" or a "Reflective Questioner Agent").
Goals & Outcomes: Clearly specifies the desired user outcome for that module (e.g., "User leaves with 3 concrete action steps").
Methodologies: Specifies the coaching framework the agent must adhere to (e.g., GROW Model, SMART Goals, or Cognitive Behavioral Coaching principles).
Key Question Bank: Contains the specific, targeted questions designed to elicit the required information for the module’s success.

This modularity ensures precision, quality control, and faster iteration of coaching techniques.

6. Leveraging LLMs as Micro-Services for Advanced Text Processing

While Rocky.ai’s core systems are proprietary, the platform strategically integrates the power of Large Language Models (LLMs) as internal micro-services to enhance text processing capabilities. Rocky.ai remains agnostic in its use of providers, leveraging state-of-the-art solutions such as OpenAI’s ChatGPT, Google Gemini, and Anthropic Claude to deliver top-notch text summarization and analysis.

As a large-scale business account with these providers, Rocky.ai is opted out of any training and data collection practices, ensuring that user data and intellectual property are never shared with or used by these systems according to GDPR.

These LLM services are exclusively used for processing text fragments generated by Rocky.ai’s proprietary AI, ensuring secure, high-quality outputs without compromising user data.

Anonymization and Pseudonymization: All user conversation data, insights, and derived memory structures are processed using advanced anonymization and pseudonymization techniques. This ensures that data utilized for model training, system improvement, and analysis cannot be directly linked back to an individual user's personal identity, maintaining a high standard of privacy compliance.
Ethical AI Use: Rocky.ai's commitment to ethical AI development extends to the handling of sensitive self-coaching data, guaranteeing that user input is utilized solely for enhancing the personalized coaching experience and improving the efficacy of the proprietary AI models.