Life sciences
Can every prediction about a medicinal plant be traced to a compound, a target and a study?
Research
Four areas, one question
How do you know an AI system's output is right? We study that question where being wrong is costly: medicine, the news, generated media and software.
Realtime video
Can generated video keep up with the display and stay consistent over time?
Media intelligence
Can AI track what the press says without inventing what it said?
Software from intent
Does a generated app do what was asked, and how would you know?
Why these four
The areas look different. The problem underneath them is the same. A model produces something fluent: a ranking of drugs, a video, a summary of the news, a working app. Fluency is easy to see. Whether the output is right is not.
In each area we build the model, build the evaluation that checks it, and publish both. Where the work is ready to use, OneThought AI turns it into a product, and what we learn from real use comes back to the research.
How each area reports
Every area page follows the same order: the question, why it matters, our approach, the status today, what we have found, and what comes next. Results are labelled internal until they have been reproduced outside the lab.