From Messy Data to HIPAA-Compliant AI
How Haven Headache & Migraine Center Built an LLM Strategy in Just Two Weeks
Haven Headache & Migraine Center
SUMMARY
Haven Headache & Migraine Center is creating virtual-first headache and migraine care, based around a daily headache-tracking text message system.
Haven knew LLMs could be helpful for their daily text check-in concept, but with so many options out there, they were uncertain where to direct their limited bandwidth.
The “messy data” of text message chains and a need to be HIPAA-compliant presented particular challenges that prompted Haven to reach out to Hop for guidance.
Hop performed a manual review of Haven’s dataset, sparking ideas about how LLMs could help, as well as a quick non-LLM win that achieved 98.6% accuracy for one of their core use cases.
In just two weeks, the Haven Headache team had a clear LLM roadmap – a sense of when to use LLMs, and which ones to use.
THE COMPANY
Haven Headache & Migraine Center is creating virtual-first headache and migraine care designed to put patients in the driver's seat. The company stems from personal experience and passion – Haven's CEO, Izac Ross, has grappled with chronic migraine since he was two. A key element of Haven's virtual-first approach is a daily headache-tracking text message, which allows Haven to monitor a patient's progress and make adjustments to their plan. As Haven grows, the daily tracking will also enable Haven to create a proprietary, first-of-its-kind anonymized headache dataset that can be useful for future population health-level research and interventions.
THE CHALLENGE
Haven's CTO, Matt Nunogawa, runs a small engineering team of three. He knew LLMs could be helpful in both the short term and long term for their daily text check-in concept, but he wasn't certain where he should direct his engineers' very limited bandwidth first. With so many LLMs available, choosing the right one is more important than ever. Matt hoped Hop Labs could quickly test amongst viable LLM options by comparing cost, accuracy, and speed.
“I’ve been writing code for decades, but the state of AI has been advancing so fast that it’s hard to keep up. I can definitely count on Hop to be on top of things like budgets and cloud compute for AI projects.”
Additionally, Haven's headache-tracking dataset consisted of unstructured text message chains – exactly the kind of "messy data" that many companies believe makes them unprepared for AI. It was clear to Matt that they needed expert guidance before proceeding with an LLM implementation.
Because Haven is a healthcare provider, they faced the added challenge of needing their LLM applications to be HIPAA compliant. They had been engaging with support teams from some of the large LLM providers, but were having difficulty getting a BAA signed. Haven hoped that if they could use smaller, open-source models for their needs, they could ensure HIPAA compliance by running everything on their own tech stack.
THE APPROACH
As a first step toward building a strategy for Haven, the Hop Labs team did a manual review of their dataset. This sparked ideas around how LLMs could help, and even unearthed a number of quick wins that could be implemented without LLMs. For example, a "fuzzy matching" algorithm yielded 98.6% accuracy for identifying and extracting the correct treatment from the text message threads.
Rather than recommending a generic AI solution, our in-depth review of Haven's specific patient communication patterns revealed unique opportunities. This deep analysis of their actual data enabled us to propose tailored solutions that addressed Haven's specific challenges in headache care management.
“In my head, LLMs are big and expensive and do a lot of amazing stuff. But for some of our use cases, Hop was able to point out shortcuts that I can just put my engineers on now to make our system smarter, without necessarily having to invest a lot of time, energy, or money.”
While Haven's team was technically capable, navigating the rapidly evolving LLM landscape would have required substantial time investment from already stretched-thin engineers. Hop’s specialized knowledge allowed rapid evaluation of multiple options and clear, actionable recommendations that could have taken Haven months to develop independently.
Our primary near-term recommendation for Haven was to leverage LLMs for creating a structured database from their text threads (i.e., "conversation normalization"). To explore this, we tested the latest open-source LLM models to see if they would meet Haven's needs. We started from the smallest – first testing 1B- and 4B-parameter models. Neither of these seemed sufficient, but a 12B-parameter model from the Gemma 3 family showed promising results. This meant that Haven would be able to use a relatively small, inexpensive LLM model in a fully compliant manner to structure their database.
Haven’s HIPAA compliance requirement eliminated many off-the-shelf LLM solutions. We specifically evaluated models that could run within Haven's existing secure infrastructure, ensuring patient data never left their protected environment. This approach eliminated the BAA complications they had been experiencing with larger LLM providers while maintaining full compliance.
THE RESULTS
In just two weeks, the Haven Headache team had a clear LLM roadmap – a sense of when to use LLMs, and which ones to use.
We identified some near-term quick wins, such as the fuzzy matching algorithm that achieved 98.6% accuracy for one of their core use cases.
Our recommendation of the 12B-parameter Gemma 3 model provided an optimal balance of performance and efficiency, reducing Haven's projected implementation costs by approximately 80% compared to larger models, while still achieving the accuracy they required.
Completing the entire engagement, from initial data review to final recommendations, in only two weeks allowed Haven to begin implementation immediately rather than spending months evaluating options.
“Everything I wanted out of the project I certainly felt was achieved. I have a lot of direct guidance and scoping on the short-term achievables as well as a better sense of how we could stack these for longer-term use cases. Plus, I’m clearer on how much things cost and where we can invest for the best ROI.”
With a clear LLM roadmap in place, Haven is now positioned to systematically enhance their patient care through AI implementation. Starting with immediate wins and building toward more sophisticated applications – all while maintaining strict HIPAA compliance – they’ll be able to optimize their limited engineering resources to reach the next level of innovation on their path.
Looking for the best ROI for your team when it comes to LLMs? Contact us to learn how Hop can help.