Clinical documentation · Production

Included Health builds Wordsmith, an internal GenAI platform powering healthcare documentation, RAG, and agentic workflows

The problem

Included Health's data science team needed to modernize NLP capabilities last advanced in 2021, and care coordinators were spending time on manual documentation after every member encounter. The team also lacked a unified, secure way to route LLM requests across cloud providers in a HIPAA-regulated environment.

Workflow diagram · grounded in source
1
Member interaction triggers workflow
trigger
“speeding up manual documentation that care coordinators were writing after each encounter with a member”
2
LLM request routed via proxy
routing
“all LLM requests at Included Health pass through a single internal service called wordsmith-proxy, which then routes them to the appropriate provider”
3
Call and chat transcription
ai_action
“now covers both chats and calls, using transcription capabilities via a Whisper model hosted on Wordsmith Serving”
4
Ghostwriter documentation output
output
“This project automatically generates documentation for care coordinators after member interactions”
5
RAG insurance document retrieval
ai_action
“answering insurance plan questions by retrieving relevant documents”
6
Confluence documentation query
integration
“an agent in ChatIH that can query for internal Confluence documentation, which is now used by over 80 of our engineers”
7
Clinical Scribe SOAP note generation
output
“It supports real-time visit transcription and generation of medical documentation, including SOAP notes”
Reported outcome

Wordsmith was deployed across multiple production applications: ChatIH reached 400 internal users, a Confluence agent was adopted by over 80 engineers, Ghostwriter automated care coordinator notes, Coverage Checker launched to the first external customer, and Clinical Scribe automated clinical documentation.

Reported metrics
ChatIH internal users400
Engineers using Confluence agentover 80
Care coordinator documentation speedspeeding up manual documentation
Healthcare provider efficiencyenhance healthcare provider efficiency
Reported stack
GPT-4OpenAI Python SDKGoogle VertexAIAWS BedrockAzure OpenAIMLServerHuggingFaceMLFlowWhisperKubernetesKarpenterLlamaIndexConfluenceGoogle VertexAI, AWS Bedrock, Azure OpenAI
Source
https://includedhealth.com/blog/tech/building-a-platform-for-genai-at-included-health/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Wordsmith was deployed across multiple production applications: ChatIH reached 400 internal users, a Confluence agent was adopted by over 80 engineers, Ghostwriter automated care coordinator notes, Coverage Checker la…

What tools did this team use?

GPT-4, OpenAI Python SDK, Google VertexAI, AWS Bedrock, Azure OpenAI, MLServer, HuggingFace, MLFlow, Whisper, Kubernetes.

What results were reported?

ChatIH internal users: 400; Engineers using Confluence agent: over 80; Care coordinator documentation speed: speeding up manual documentation; Healthcare provider efficiency: enhance healthcare provider efficiency (source-reported, not independently verified).

How is this clinical documentation AI workflow structured?

Member interaction triggers workflow → LLM request routed via proxy → Call and chat transcription → Ghostwriter documentation output → RAG insurance document retrieval → Confluence documentation query → Clinical Scribe SOAP note generation.

Related clinical documentation cases