Ryght’s Journey to Empower Healthcare and Life Sciences with Expert Support from Hugging Face

Life sciences companies sit on mountains of data — lab results, EMR records, genomics, claims, pharmacy logs — yet turning that pile into usable insight still takes armies of analysts and painfully slow workflows. Ryght, a startup building enterprise AI copilots for this exact industry, is trying to fix that gap. This week it opened Ryght Preview to the public, a free platform aimed at researchers and knowledge workers who need answers faster than a team of PhDs typing SQL queries.

What's more interesting than the launch itself is how Ryght got here. Rather than going it alone, the company brought in Hugging Face early through its Expert Support Program, essentially hiring a technical advisory team to help navigate an AI landscape that changes weekly. CTO Johnny Crupi credits that partnership with speeding up development significantly, singling out Hugging Face's Text Generation Inference (TGI) and Text Embeddings Inference (TEI) tools as central to the build.

The collaboration solved three real problems. First, keeping a small team current on fast-moving ML techniques without drowning in noise — Hugging Face's experts filtered signal from hype through workshops and regular advisory calls. Second, choosing which of the countless available models and frameworks actually fit life sciences use cases, rather than chasing whatever's trending on Twitter that month. Third, and probably the trickiest, building an architecture secure and flexible enough for enterprise healthcare clients while still being able to swap in new medical LLMs as they appear.

That last point shows up concretely in how Ryght structured its system. Each LLM gets registered against customer-specific inference endpoints, so the platform can plug in a new specialized medical model without ripping out existing infrastructure. Pairing TGI with TEI let Ryght move off proprietary embedding APIs entirely, cutting latency and dodging rate limits while running fine-tuned embedding models suited to biomedical text specifically. Behind the scenes, batching and GPU distribution keep the system responsive even when multiple enterprise customers hit it at once.

None of this guarantees Ryght wins in a crowded healthcare-AI market, but the approach — lean on open infrastructure, stay model-agnostic, treat flexibility as a feature rather than an afterthought — is a sensible bet in a field where today's best medical LLM is next year's outdated choice.