Since October 2025
Answering a single biological question can take weeks — hunting down datasets, decoding metadata that no two labs annotate the same way, correcting for batch effects, and writing QC and analysis code from scratch. The biology is rarely the bottleneck. The data wrangling is.
Sand does the wrangling. Describe what you want to know, and it locates the right public data, harmonizes and quality-checks it, runs the analysis, and hands back a clear answer — together with the full code, data, and step-by-step reasoning. Not an answer you have to trust. One you can verify.
Public biology data has exploded — millions of datasets sit in open repositories, mostly unused because they're too messy to combine. AI can finally read metadata, harmonize studies, and write analysis code well enough to unlock them. The tool that never existed can finally exist — and, crucially, show its work.
Anyone who works with biological data knows the feeling: you have a clear question — which sites drift with age in the liver, how a pathway shifts under a drug — and you know the data to answer it already exists, scattered across public repositories. Getting to the answer means weeks of unglamorous work before you can even start thinking about the biology.
I hit this wall over and over building bioinformatics pipelines. Tracking down datasets, deciphering inconsistent metadata, fighting batch effects, writing the same quality-control code again. The biology was never what slowed me down — the plumbing was. And the answer was almost always sitting in public data, waiting for someone to do the careful, tedious work of getting it out.
For a while the promise was that an AI chatbot would answer these questions. It can't — not reliably. A confident, plausible, wrong answer is worse than useless in science, and generic tools hide their reasoning, so you can never check them. Sand takes the opposite stance: it does the wrangling and shows every step — the datasets it chose, how it harmonized them, the QC it ran, the code, the reasoning — so the answer is reproducible by construction.
The hardest part of working with public biological data is the metadata — no two labs annotate a study the same way, and nothing downstream works until it's reconciled. We've built a pipeline that harmonises it, the piece most tools skip because it's slow, unglamorous, and difficult. It took four months to get right and it is mainly for methylation arrays extracted from the gene expression omnibus GEO.
We're now building the layer on top: a universal quality-control step the user controls and directs to their own analysis — so the same engine adapts to each researcher's standards instead of forcing one rigid pipeline on everyone. Correctness before coverage: every capability has to produce a result a domain expert can check and stand behind before it ships.
The research tool is a real business on its own — research teams and biotech will pay for reproducible answers they can trust. But it is step one of a deliberate path toward personalized medicine as a closed ecosystem: predicting the right intervention for each individual, then designing and supplying the compounds made for them. Every question the engine answers builds the same foundation — accessing, harmonising, and reasoning over the world's biological data correctly — and each rung compounds the data and the trust the next one needs.
The long-term goal of Sand is to design and supply medicine made for one person at a time — the right drug, or combination, matched to each individual's biology. We get there in steps: first a data engine that turns the world's public biological data into reproducible answers, then predicting how the body responds to an intervention, then prescribing the right combination for each person, and finally producing the compounds themselves — a closed ecosystem, from data to medicine.
We start by making biology's existing knowledge usable — and aim to end with medicine designed for one person at a time.
Sand first competes with the way biology teams currently answer questions from public data: internal bioinformatics teams, custom scripts, consultants, and platforms such as DNAnexus, Terra, and Seven Bridges that help researchers manage data and run workflows. These tools are powerful, but they are still mostly workflow platforms.The user needs to know what data to use, how to clean it, what analysis to run, and how to interpret the result.
Sand is different because it starts from the biological question itself and returns a reproducible answer: the datasets used, the harmonization steps, the quality checks, the code, and the reasoning. The first wedge is not “AI drug discovery.” It is verifiable biological analysis from public data. Over time, that same engine becomes the foundation for predicting intervention response, selecting the right treatment combination for each person, and eventually designing and supplying personalized compounds.