Loka Brings Claude to Mental Health Care, HIPAA Intact
Industry
Healthcare, AI/ML, Editorial/Content Creation
Tech & TOOLS
Amazon Bedrock, Anthropic Claude, Amazon SageMaker, LangChain, AWS Lambda, Amazon S3
Teams & Services
AI Engineering, DevOps,
Project Manager
milestones
0-PoC in six weeks
HIPAA Compliance
Industry
Healthcare, AI/ML, Editorial/Content Creation
Tech & TOOLS
Amazon Bedrock, Anthropic Claude, Amazon SageMaker, LangChain, AWS Lambda, Amazon S3
Teams & Services
AI Engineering, DevOps,
Project Manager
milestones
0-PoC in six weeks
HIPAA Compliance
One in five Americans suffers from mental illness. Dr. Katz is leveraginge Claude on Amazon Bedrock to amplify wellness therapy, meeting the scale of need.
The goal
Dr. Katz offers a personalized patient engagement platform that uses video to transform care, helping providers focus on diagnosis, communication and improvement. The startup partnered with Loka to undergo Loka’s GenAI Workshop, hoping to uncover scaling strategies and unlock efficiencies in app design and deployment, all based in generative AI.
The challenge
Dr. Katz enlisted Loka to build generative AI services that facilitate clinician-patient collaboration and clinical professional development, simplify video publishing and automate routine clinical tasks. Additionally, all system architecture must comply with HIPAA privacy regulations.
The first use case the team aimed for was a streamlined video-publishing workflow that would reduce administrative burden, save valuable time and improve quality control.
The solution
Loka stood up a working pipeline fast, then let the results pick the model. The team began with Llama 2 on Amazon SageMaker to prove the workflow end to end, benchmarking output quality against a small set of real clinical content samples. Anthropic's Claude on Amazon Bedrock won on the criteria that mattered: quality of generated titles, descriptions and tags, cost at scale, and HIPAA eligibility. Loka's AWS-certified ML engineers built the production pipeline around it.
"From our very first conversations with the Loka team,” said Nathaniel Hundt, Dr. Katz founder and CEO, “they were extremely interested in the work and the mission of the company and would roll up their sleeves alongside us to figure out the best possible solutions.”
Hundt backed the recommendation and the architecture reflects his longer-term intent. LangChain handles prompt construction and model orchestration, which keeps Dr. Katz portable. As stronger models ship, the team can adopt them without rebuilding the pipeline. "We want the best available model for our use cases, now and in the future," Hundt said.
What we delivered
Ever mindful of budget, Loka configured AWS cost controls such as budget alerts and infrastructure pausing during inactivity.
“We find that this is something that we’re able to deploy because we can afford it, and that matters to us,” Hundt said. “We want to roll generative AI into our application in a manner that’s not cost prohibitive.”
To create a scalable system, we built a workflow responsible for fetching transcriptions from an S3 bucket and providing LLM-based suggestions. This workflow is handled by a Lambda function, which is triggered when a transcription file is uploaded to an S3 bucket. Subsequently, it sends requests to the SageMaker endpoint hosting the LLM, which in turn, provides suggestions for titles, descriptions and tags.
- Deployed LLMs on Amazon SageMaker
- Efficient use of LangChain for LLM workflow handling with Llama 2
- AWS budget alerts and infrastructure pausing curb cost
- Scalable workflow fetches transcriptions from S3 bucket
- Lambda function integration triggers enhancements via Amazon SageMaker
Project highlights
80% less time to publish
Claude on Amazon Bedrock generates titles, descriptions and tags the moment a transcript lands in S3. Metadata that took clinicians roughly 15 minutes per video now takes under 3, review included.
HIPAA Compliance
Claude is HIPAA-eligible through Amazon Bedrock, so Dr. Katz never traded output quality for compliance. Loka configured the surrounding S3, Lambda and SageMaker pipeline for PHI handling from day one, not as a pre-launch retrofit.
100% publishing with metadata
Every video ships with a title, description and tag set. Consistency that depended on individual clinicians is now automatic.
40% lower infrastructure cost
Budget alerts and automatic pausing during inactivity cut monthly AWS spend, with further savings from Bedrock's pay-per-token pricing over always-on endpoints.
”We've enjoyed every interaction with the Loka team. We found that they have skilled leaders who operate on the product side and understand the direction of the market through their relationships with other leading customers.”
.avif)
