ViewOps User Memory Architecture

ViewOps uses a hybrid memory model to track and personalise user interactions across SQL-based queries. The goal is to make answers more relevant, secure, and consistent — without relying solely on the language model.

Structured Memory (PostgreSQL)

Core user data is stored relationally to support:

• Access Control — what data each user can see (e.g. hide financials for Livestock team)
• Prompt History — what they asked, how it was answered, and what tools were used
• Tag Context — links each user to departments, topics, metrics, and ownership
• Feedback Loops — records if the answer was correct or corrected by the user
• Edit Tracking — captures how often users adjust AI output or trigger manual overrides

Semantic Memory (Vector DB – e.g. pgvector)

Vector embeddings are used to semantically match:

• New questions to previous ones
• Fields/metrics to user context
• Prompts to appropriate predefined tools

This allows the system to auto-suggest tools, detect ambiguity, or route queries with higher confidence.

Example Use Cases

• “What about Abattoir 2?” → understood from previous batch query as a supplier site
• User who always checks Yield Variance → answer formatted as table + chart