Avatar Platform — Trulience

The avatar is built on Trulience and serves as the primary user-facing interface. Three AI components are configured within Trulience to handle conversation, input, and output:

• Language model: Llama, hosted on Groq
• Speech-to-text: Deepgram — transcribes user speech to text in real time, chosen for accuracy and cost efficiency
• Text-to-speech: ElevenLabs — generates the avatar's spoken responses using a Caribbean female voice, regarded as the industry standard for voice quality

Trulience homepage — the interactive avatar platform powering CaribDX

Trulience Avatar Creator — where appearance, brain, and voice components are configured

Function Calling & Workflow Trigger

When a user asks the avatar for a Caribbean health update, a function attached to the avatar fires a webhook call into an n8n automation workflow. This workflow handles data retrieval and story delivery, returning the response through to the avatar for playback on the front end.

n8n workflow — CaribData News Aggregator: Webhook → Static Stories → Writing AI Agent → Respond to Webhook

Data Layer — Supabase

Caribbean health statistics data is stored in a Supabase open-source database. The schema contains multiple tables, with the core health dataset forming the primary source for story generation.

Supabase — the open-source Postgres platform hosting the CaribDX health dataset

Schema visualiser — showing the three tables: datasets, unwpp_population, and hba1c_synthetic

Table editor — unwpp_population table with 2,646 records of Caribbean demographic health data

Data Analysis & Story Generation Pipeline

The pipeline from raw data to avatar-ready stories runs as follows:

• A Python function performs data discovery and analysis against the Supabase database
• Claude generates narrative health stories from the analysis output
• Generated stories are injected via API into the n8n workflow as static story content
• When a user query arrives, n8n retrieves the relevant story and returns it through to the avatar

Execution logs within n8n provide a full historic record of workflow runs and story outputs.

CaribDx Automated Data Intelligence Pipeline — three-phase process: data discovery (Supabase), AI story generation (Claude Sonnet), and n8n workflow injection

End-to-End Flow