Sempre Health, a startup tackling medication adherence through SMS engagement and prescription discounts, just hit a milestone that most ML teams spend years chasing: their production NLP pipeline now automatically handles nearly 20% of all inbound patient messages. That’s a significant chunk of work that previously landed in the patient operations team’s ticket queue.
Before working with Hugging Face’s Expert Acceleration Program, Sempre Health relied on a rule-based system that caught about 80% of incoming texts. The problem? Rule-based parsing plateaus fast. Swaraj, the company’s CTO and co-founder, knew a statistical machine learning approach was the only path forward, but he and ML engineer Larry were unsure whether they were formulating the problem correctly or picking the right model architecture for their specific training data.
The Hugging Face team did more than recommend models. According to Swaraj, they provided guidance on getting representative and accurate labels for text messages — a step that quietly makes or breaks NLP projects — and saved the team ‘countless hours of research time’ by pointing them immediately to the right methods. What stood out to Swaraj was the lack of friction: the experts ‘met us where we were’ and didn’t pile on extra work.
The result is a system that can distinguish a simple ‘Thank you’ — which gets an automated ‘You’re welcome’ — from a ‘Can you refill my prescription?’ — which triggers an automated pharmacy call and refill request. That second use case is the interesting one. It’s not just classification; it’s classification wired directly into a real-world action. For a small team just starting to explore ML, that’s the difference between a proof-of-concept and something that changes daily operations. The company’s recommendation for the program is aimed squarely at teams that want AI pipelines in production without the typical development grind.
