Virtual AI Assistant for Alzheimer's Foundation of America.

The Alzheimer’s Foundation of America approached us to build a sophisticated AI assistant that could support their community by answering questions about Alzheimer’s, dementia, and brain health. We built Allison. The production system she powers is documented in Case File 001: the AFA AI Assistant Powered Helpline; this is the story of how she was made.

AI avatar creation.

We worked meticulously to design Allison, generating the features, gender, hairstyle, clothing, and setting to align with AFA’s needs. Roughly 300 images were generated, with about 150 options presented to AFA for feedback. The process combined text-to-image generation, inpainting, and facial swapping using a custom-trained Stable Diffusion XL model merged with additional portrait training. Final touches included inpainting a New York bokeh background to reflect AFA’s Manhattan location and adding an AFA badge.

AI agent and conversational AI development.

Allison uses vector databases and OpenAI’s API to convert text into embeddings and provide natural language responses. While she excels at answering questions, Allison is designed to recognize her own limitations: the system can create helpline tickets that connect users with AFA social workers through secure forms.

Functional AI development.

Pydantic classes standardize output formatting, which enables Allison’s reliable agentic behavior and even helps with decision making. This is the unglamorous layer that makes conversational AI dependable enough for a healthcare nonprofit.

Natural user interface.

We partnered with D-ID for streaming avatar functionality. The workflow: user input, agentic decision making, OpenAI processing, then D-ID video streamed over WebRTC to the browser. ElevenLabs integration enabled custom voice selection.

AI social assets.

Promotional materials were produced with D-ID studio features, ComfyUI workflows, and LivePortrait tools.

AI agents and avatars are redefining digital experiences. Large language models play a crucial role in making these systems functional, engaging, and adaptive to user needs.