LLM-powered Real-time Support With Safety and Continuity Built-in
We built a secure AI coaching engine designed to deliver real-time, emotionally aware support through consistent, guideline-compliant conversations on personal devices.The platform was created to make behavioral coaching more accessible and scalable by allowing users to interact with AI personas such as a coach, friend, or “big brother” figure that could provide encouragement, accountability, and context-aware support over time.
Ptera implemented the AI infrastructure powering these interactions, combining persona-driven LLM workflows, contextual memory recall, and safety-focused response validation to support more natural and supportive conversations at scale.
Challenge & Strategy:
- Provide empathetic, context-aware behavioral-coaching chats at scale following motivational-interviewing best practices.
- Deliver safe, high-quality real-time conversations that retain long-term context through specialized AI expertise and infrastructure.
Technical Approach & Integration
- Fine-tuned a GPT-based LLM to simulate multiple supportive personas (coach/friend/“big brother”).
- Prompt templates maintained MI coaching tone; RAG (vector DB) personalised context recall.
- Rules-based safety layer (content filtering + emotional-support heuristics).
- Deployed as a secure microservice-chat frontend calls the AI service; all sessions logged; memory management ensures continuity.
