MongoDB Data Models

Relevant source files

The following files were used as context for generating this wiki page:

• Ai-Assistant-v2/auth-service/models/User.js
• Ai-Assistant-v2/auth-service/models/Conversation.js
• Ai-Assistant-v2/auth-service/models/Memory.js
• Ai-Assistant-v2/auth-service/server.js
• Ai-Assistant-v2/auth-service/routes/ai.js

The AI Assistant platform utilizes MongoDB as its primary persistence layer to manage user identities, complex conversation histories, and a long-term memory system. These models are defined using the Mongoose ODM (Object Data Modeling) library, which provides schema validation and relationship mapping between documents.

The data architecture is designed to support a multi-user environment where each user's data is isolated. Key functionalities such as JWT-based authentication, vector-based memory retrieval, and multi-model AI switching (Phi-3 and Gemini) rely directly on the structure of these models.

Sources: README.md, Ai-Assistant-v2/auth-service/server.js:47-51

Entity Relationships

The data model follows a relational pattern within a NoSQL structure, primarily centered around the User entity. Both Conversation and Memory documents maintain a reference to the User via an ObjectId.

Detailed Schema Definitions

User Model

The User model handles authentication credentials and global application preferences. It includes a nested settings object that dictates the behavior of the AI interface for that specific user.

Field	Type	Description	Constraints
username	String	Unique identifier for login	Required, Unique
email	String	User's email address	Unique, Sparse
password	String	Hashed user password	Required
settings.defaultModel	String	Preferred AI engine	Enum: ['phi3', 'gemini'], Default: 'phi3'
settings.theme	String	UI appearance	Enum: ['dark', 'light'], Default: 'dark'
settings.memoryEnabled	Boolean	Toggle for memory retrieval	Default: true

Logic & Middleware:

• Password Hashing: Uses a pre('save') hook to hash passwords with bcrypt before persistence.
• Verification: Includes a comparePassword instance method for authentication checks.

Sources: Ai-Assistant-v2/auth-service/server.js:29-51, Ai-Assistant-v2/auth-service/models/User.js:4-21

Conversation Model

The Conversation model stores the full transcript of interactions. Rather than storing messages in a separate collection, they are managed as an array of sub-documents to maintain chronological integrity and simplify retrieval for context-aware AI prompts.

Field	Type	Description
userId	ObjectId	Reference to the owner
messages	Array	List of message objects (role, text, timestamp, intent, sentiment)
model	String	The AI model used for this specific thread
updatedAt	Date	Timestamp of the last interaction

Indexing:
The model uses a compound index on { userId: 1, updatedAt: -1 } to efficiently fetch a user's most recent chat threads.

Sources: Ai-Assistant-v2/auth-service/models/Conversation.js:3-48, Ai-Assistant-v2/auth-service/routes/ai.js:49-52

Memory Model

The Memory model is the core of the platform's "Intelligent Memory System." It stores distilled facts extracted from conversations, enhanced with vector embeddings for similarity searches.

const MemorySchema = new mongoose.Schema({
    userId: { type: mongoose.Schema.Types.ObjectId, ref: 'User', required: true, index: true },
    fact: { type: String, required: true },
    sourceConversationId: { type: mongoose.Schema.Types.ObjectId, ref: 'Conversation' },
    importanceScore: { type: Number, min: 0, max: 1, default: 0.5 },
    tags: [{ type: String }],
    embedding: { type: [Number], default: null },
    createdAt: { type: Date, default: Date.now }
});

Sources: Ai-Assistant-v2/auth-service/models/Memory.js:3-31

Data Flow: Chat and Memory Persistence

When a user sends a message, the system orchestrates data across multiple models to provide a context-aware response.

Memory Management Logic

Memories are not merely stored; they are ranked and retrieved based on relevance.

1. Extraction: During chat, if the Python AI service identifies "memory candidates," the system filters them based on an importanceScore threshold (default 0.6).
2. Indexing: The Memory schema utilizes multiple indexes to optimize performance:
   • userId: 1, importanceScore: -1: For fetching the most significant facts.
   • userId: 1, embedding: 1: Facilitates vector-based similarity lookups.
3. Training: A dedicated endpoint POST /train-memory allows the system to retroactively scan the last 5 conversations, extract facts, and populate the Memory collection while avoiding duplicates via regex-based existing-fact checks.

Sources: Ai-Assistant-v2/auth-service/models/Memory.js:34-36, Ai-Assistant-v2/auth-service/routes/ai.js:155-200, Ai-Assistant-v2/auth-service/services/aiService.js:52-60

Summary

The MongoDB data layer provides the structural foundation for the AI Assistant's advanced features. By leveraging Mongoose schemas for Users, Conversations, and Memories, the system maintains a clear separation of concerns while enabling complex interactions like vector-based context retrieval and automated memory extraction. This architecture ensures that user preferences and long-term knowledge persist across sessions, significantly enhancing the conversational intelligence of the platform.