User Model

The dimensional blueprint that orchestrates memory preservation, retrieval, and interpretation for domain-specialized intelligence

The user model (L3) is more than just a data structure—it's the operational blueprint of the entire functional memory system that maintains a holistic understanding of the user across all dimensions and serves as the foundation for live reasoning. As the global user model constantly in memory during live sessions, L3 is a critical enabler of the unified Memory-Knowledge-Reasoning (M-K-R) cognitive process, providing the rich, structured Memory that informs Knowledge application and frames Reasoning. By defining dimensional priorities and relationships, L3 orchestrates how information flows through the system and is preserved, retrieved, and interpreted, facilitating the cyclical optimization inherent in the M-K-R paradigm. This interconnected approach recognizes that effective understanding requires seeing the user as a whole person rather than maintaining a myopic focus on an uncontextualized topic.

Foundation for Live-Agent Reasoning

L3's most critical function is to maintain all function-specific information (Memory) with the right interpretation at the right granularity at present time to support live-agent reasoning (Reasoning), which in turn determines how Knowledge is best applied.

1

Complete Functional Information

Having all the Memory needed for agent functions ready at session start to inform Knowledge and Reasoning.

2

Correct Contextual Interpretation

Ensuring information from Memory is interpreted within the proper domain context to guide Knowledge selection and Reasoning paths.

3

Appropriate Granularity Level

Providing Memory at the right level of detail—neither too general nor too specific for the current M-K-R task.

4

Real-Time Availability

Making this Memory immediately accessible without additional retrieval steps in most cases, ensuring high-bandwidth for the M-K-R interplay.

This present-time information foundation is what enables agents to reason effectively without constantly retrieving and reconstructing context. By maintaining this rich, function-optimized information state in L3, the agent can focus processing resources on the cyclical interplay of Memory, Knowledge, and Reasoning, rather than basic information gathering.

For domain-specialized agents, there exists a critical cluster of information within L3 that must all be present at the perfect interpretation and depth to inform all aspects of the M-K-R cycle. This information cluster plays a vital role in everything from filling relevant information gaps (Memory influencing Knowledge/Reasoning) to applying domain knowledge (Knowledge powered by Memory, shaping Reasoning), recontextualizing past interactions (Knowledge/Reasoning updating Memory), and guiding reasoning patterns and explicit reasoning (Reasoning drawing from M&K). Without this perfectly calibrated information foundation in L3, specialized agents would be unable to leverage their domain expertise effectively, as they'd lack the contextual architecture necessary for sophisticated M-K-R integration.

For example, if a patient goes to a physician complaining about jaw pain, it may be relevant to consider their history of heart conditions, as this can be a non-traditional symptom of a heart attack. This cross-domain correlation capability is essential for medical intelligence performance—the system must organize high-dimensional personalized data to maintain awareness of how seemingly unrelated symptoms connect to established risk factors.

Our user model maintains contextual awareness across clinical domains, enabling intelligent connections between disparate information points that might otherwise remain isolated. This organized approach to complex patient data allows the system to identify subtle correlations that are critical for accurate diagnosis and treatment decisions, transforming medical AI from simple information retrieval into true clinical intelligence.

Why User Models Matter

• ✅ Guarantee Critical Recall: Ensure life-or-death information is never forgotten

• ✅ Optimize Resources: Concentrate computational resources where they deliver highest value

• ✅ Reduce Errors: Address the root cause of catastrophic errors in decision systems

• ✅ Functional Alignment: Adapt memory to specific industry needs and use cases

• ✅ Contextual Coherence: Maintain perfect context for information that requires it

• ✅ Information Evolution: Track how understanding changes over time with proper recontextualization

• ✅ Dimensional Awareness: Organize information by functional importance for intelligent retrieval

• ✅ Real-Time Reasoning Support: Provide the complete information foundation needed for agent reasoning without delays

Dimensional Framework in Action

Each dimension in the user model defines a specific category of information with associated precision requirements and contextual preservation needs:

{
  "description": "Medical & Health History: Current and past health conditions, hormonal and metabolic profiles, treatment experiences, and medication adherence that provide context to the client's physical wellbeing.",
  "tags": ["health", "clinical", "medical history"],
  "precision_required": "perfect"
}

Core Capabilities

This dimensional structure allows the system to:

1. Define perfect recall boundaries: Clearly establish which information must never be forgotten or incorrectly contextualized.

2. Prioritize computational resources: Allocate memory resources based on functional importance.

3. Guide context preservation: Maintain complete contextual relationships for critical information.

Operational Benefits

1. Inform retrieval operations: Focus search near known important information.

2. Structure user understanding: Organize information by functional relevance rather than arbitrary categories.

3. Support real-time reasoning: Ensure all function-critical information is readily available at the right granularity.

Enabling Critical Enterprise Capabilities

Conditional Dynamic Behavior Selection

The user model enables behavior routing based on dimensional state.

For example, for a sample user named Tony:

When discussing workout options:
- User model indicates past ACL, rotator cuff, knee injuries
- System automatically routes to "Injury-Conscious Exercise" behavior
- Modifies recommendations without needing to retrieve full injury details

This immediate conditioning on dimensional understanding happens without any additional retrieval operations in 90% of interactions.

Accelerated Personalization

Since the user model provides complete dimensional understanding at session start, personalization happens instantly in most cases:

Tony: "What should I eat today?"

Without user model: Generic response requiring multiple follow-ups about preferences, restrictions, and goals.

With user model: Immediate response incorporating:
- High-protein dietary approach
- GLP-1 medication side effect considerations
- Strategies to manage binge eating tendencies
- Quick meal options fitting busy schedule

This instant personalization occurs without additional retrieval steps in approximately 90% of interactions.

Intelligent Expansion Query Generation

When information gaps exist, the user model generates expansion queries that reference existing dimensions for super-efficient retrieval:

Current Session: Tony mentions morning energy crashes
Expansion Query: "How does um:energy_patterns relate to um:meal_timing and um:medication_schedule?"

Current Session: Patient reports new chest discomfort  
Expansion Query: "How does um:cardiac_history connect to um:stress_patterns and um:exercise_changes?"

These expansion queries are super efficient because they reference established user model dimensions (um:dimension_name) rather than requiring standalone searches. They bridge current session concerns with historical context through the complete dimensional framework.

Targeted Deep Reasoning over Raw Conversations

When the system accesses L0 data (raw conversations), the user model provides essential interpretive context:

Raw mention: "I'm feeling tired in my leg today"

Without user model: Generic concern about fatigue.

With user model: Interpreted through injury history dimension as potential injury-related fatigue requiring careful monitoring.

This dimensional interpretation transforms raw data into functionally valuable insights that serve the specific enterprise need.

Reconciling Contradictory Information

The user model enables sophisticated resolution of conflicting information through dimensional weighting:

In medical contexts: Recently reported symptoms receive higher weight than historical self-reports.

In Tony's case: Recent factual statements about his eating behavior would outweigh older statements about dietary preferences, while core medical information maintains consistent high priority.

This dimensional approach to contradiction resolution ensures functional consistency rather than simplistic "most recent wins" approaches.

Cross-Session Continuity

By maintaining dimensional understanding across sessions, the user model enables perfect continuity:

Session 1: Tony mentions starting GLP-1 medication
Session 5: Tony reports side effects
Session 12: System seamlessly references both the medication and side effects when discussing nutrition without requiring explicit recall

This continuity happens automatically because the user model maintains dimensional relationships between critical information elements.

Implementation Impact

Healthcare

Enables consistent treatment continuity across provider changes

Financial Services

Maintains perfect compliance information across advisor transitions

Legal

Preserves case precedent relationships with complete contextual understanding

Customer Support

Eliminates the need to repeat critical preferences or history

The Blueprint for Intelligence: The user model is the blueprint that ensures the memory system does not just remember facts—it understands their functional importance within the broader M-K-R cognitive process, maintains their proper relationships, and delivers them with perfect context when they matter most for Knowledge application and Reasoning. Most critically, it ensures that all function-specific information with the right interpretation at the right granularity is available in real-time as the foundation for live-agent reasoning within the unified M-K-R system.

This dimensional framework also enables Amigo's personalized evaluation system. Rather than measuring AI performance against generic benchmarks, our metrics adapt to each user's complete context, creating assessment criteria that reflect actual individual value rather than abstract performance standards.