At the heart of Tibb lies a system-level AI architecture that combines multiple models to deliver safe, culturally sensitive, and effective wellness support.
1. Understanding & Personalization
- Our AI models are trained to understand user needs, both short-term and long-term, and to match users with culturally appropriate coaching.
- Conversation feature extraction allows us to classify topics into meaningful categories for tailored support.
2. Evidence-Based Coaching
- Downstream from classifiers, LLMs generate personalized coaching grounded in clinically validated methods such as Cognitive Behavioral Therapy (CBT), with adaptive use of Dialectical Behavioral Therapy (DBT) and Interpersonal Therapy where relevant.
- Coaching content is structured into bite-sized, actionable practices for everyday wellness.
3. Guardrails for Safety & Focus
- Natural Language Processing (NLP) and sentiment analysis safeguard both incoming and outgoing conversations.
- Guardrails prevent inappropriate or off-topic content by steering conversations back toward wellness.
- Additional safeguards mitigate LLM “hallucinations” and maintain clinical integrity.
4. Measuring Efficacy
- We use Randomized Controlled Trials (RCTs) governed by Institutional Review Boards (IRBs) to evaluate Tibb’s effectiveness in improving well-being.
- Data from trials and user engagement inform continuous improvement of both our platform and coaching models.