Emotionally Intelligent AI System

A client project from 2025: an AI character platform for brand experiences, built around one idea -- a persona should remember how conversations felt, not just what was said. A character that forgets an emotional moment breaks the illusion faster than one that forgets a fact.

The memory engine stores conversational recall as Mistral embeddings in Weaviate, but vectors alone only answer what is relevant. At conversation time the system also writes state-aware memory traces -- emotional register, intensity, symbolic anchors -- and uses them to re-rank retrieval. Vectors find what is relevant; traces decide what matters.

Forgetting is tuned, not accidental: decay curves depend on emotional intensity, so strong moments persist while small talk fades. On top sits a symbolic interaction layer and modular personas, which keep a brand voice emotionally consistent across weeks of interaction.