AI Mental Health Coach

The AI Mental Health Coach built for Impact Suite was a first-to-market product in the mental health space — a proactive AI coach that reaches out to users rather than waiting for them to engage. Built on a self-hosted LLAMA-3 deployment with a RAG (Retrieval-Augmented Generation) architecture, the system draws on a curated knowledge base of therapeutic frameworks, crisis intervention protocols, and evidence-based mental health content to generate responses that are grounded, contextually appropriate, and clinically informed.

Six-Persona System with Eleven Labs Voice

One of the technical and product innovations was a six-persona coaching system — each persona tuned with a distinct communication style suited to different user states and needs. A user navigating a high-stress moment gets a different coaching tone than a user doing routine check-ins or working through a structured program. Each persona is voiced using Eleven Labs, giving the coach a consistent, natural-sounding audio identity. The combination of adaptive persona selection and high-quality voice synthesis made the interaction feel genuinely conversational rather than scripted.

Self-Hosted Infrastructure and RAG Pipeline

Running LLAMA-3 on self-hosted infrastructure was a deliberate choice — keeping sensitive mental health data off third-party model APIs and maintaining full control over inference behavior. The RAG pipeline, built with LangChain and Python, handles document ingestion, embedding, retrieval, and prompt construction, ensuring that every response is augmented with relevant, vetted content from the knowledge base. This architecture also made it straightforward to update the knowledge base as new clinical guidelines or product-specific content became available, without retraining the base model.