Digital Lab launching AI to more than 60 countries including US and India

GetTested.io has successfully launched its digital lab including various types of tests ranging from; Longevity, Hormones, Gut Biome, DNA and Vitamins. Combining Results with highly personalized recommendations through AI is a key success factor enabling their customers to make fact-based decisions based on results, to improve health and performance. The GetTested offering includes testing, supplements and online consultations. The company has expanded from Stockholm to North America and India, and is today selling to more than 60 countries per year and over 10,000 customers every month.

Customers expect speed and accuracy from their results — and GetTested had an in-house expert who was spending almost full time fact checking every result the LLMs produced. Growth and Customer Demand was not a problem. But repeatability and measuring quality at scale was a real blocker.

From Prototype to First-Class System

Early experiments with a single provider proved that AI could translate raw biomarker data into plain language advice, but often hung after 20 markers out of 40. Prompt tweaks, rate limit surprises, and manual QA created a lot of friction and was a problem for the momentum. One FTE had to manual QA all the results all the time and approve manually before releasing to the customer. Patrik - CEO also remarks that “GetTested wants to own the data to keep flexibility and know-how without being locked to a single model”. GetTested has a react web app through which the customers order lab tests, get their results and act on it.

Their first AI feature was to move to automated lab results. GetTested has a lot of internal feeds that describe helpful supplements that they wanted to tie together with the lab results - and put it all in natural language for the customer. The engineering team began by structuring the AI calls into small functions with dataset and few shot prompting (in context finetuning) to get high quality results. The problem with rate limits and stalling bio marker translation was solved with fallback models that are specified as an array in the AI calls. From there, they built Evals metrics to get control over quality when moving in a production environment. They’ve also exposed an API endpoint as a button for adding results that really show “what good looks like” for their experts to use, this adds the result to the dataset to create a sort of always-improving function. Adam - Lead AI Engineer explained that “Datasets together with evals and tracing works really well to get quality and insight.”

The evals API is very much in control in the hands of the engineers where they can specify their own custom evals, at the same time gaining from having the primitives readily at hand. For GetTested, this becomes a pretty good combination of DIY and solid SDK. As for the tracing, GetTested is using Opper’s built in tracing view. At one point they were pondering sending the traces to Datadog but the tracing view in Opper was eventually a solid option for them as it has a built-in path for datasets and more.

More recently, they’ve created a post sales chat bot that helps the customers when they’ve performed a lab test so that they utilize the lab results maximally to improve their health. For that, they had to give the bot some tools so that it can fetch the necessary results, gain knowledge of feeds and tie this together as a helpful assistant. They’ve done this via structured input/output, RAG and evals. “Tracing has been quite crucial for us when we’ve developed this so that we can see all input / output and build quality from there.” Adam remarks.

Success in the Customer Journey

Patrik Johansson - CEO, describes that the customer journey has gone from something with high ambitions but “half-baked” to world class: “Customers today get accurate result interpretation and recommendations based on their data, and results. They can ask our AI Health for a deeper understanding of their results and get personalized recommendations.” “My Team on the other side, can focus more on developing the business rather than checking results manually for correct interpretation. We are today comfortable with the health insights we provide to over 10.000 customers per month.”

“The customer journey is today a lot better and our customers get what they are looking for – and are more likely to re-test or recommend the services to others, increasing CLTV.

How the Engineers Describe It

It’s enlightening to speak with Adam, Lead AI Engineer at GetTested. His explanations of how they’ve worked at scale and speed with LLM providers' APIs let us understand the real pain of developing quite complex workflows when demands are high. As an engineer, that’s a tough but interesting position.

“I realized after a while that the quality of the models was never our bottleneck—measuring quality to identify problems was,”Going from realization to action is also a thing. Getting a happy path towards evals, datasets and measuring quality is not that easy.” Adam: “We had quite large prompts and quite often zero-shot multiple times to get results, then manual QA. We tried some different models, but honestly, rebuilding for different APIs wasn’t something we had time for.” What Adam alludes to in our discussion is quite clear as he is talking about going from 1 LLM zero shot and vibe coding into Proxy, Tracing and Evals in a more structured way. It’s a great story of how one can go from vibes to a more robust AI system. We also know now that they use a variety of different models for different tasks that they see them excel at.

What Patrik - CEO, has been quite clear with, is that they want to stay in control. So, retaining ownership of the logic is crucial - while still benefiting from a solid AI API. They opted for APIs that are transparent so that if they want to walk away, they can see and export all prompts, evals and datasets.

Looking Ahead

GetTested.io is piloting instant customized recommendations including life style changes and customized supplements, that attach to each Test Result. This means that the entire customer experience from start to finish becomes completely custom and tailored.

Patrik concludes: “We will continue to build the world’s biggest digital lab, and will partner with Opper going forward in this mission. At GetTested we empower each individual to take control of their own health at competitive pricing.”