|
Pienomial has launched AT0M, an artificial intelligence model that businesses can own, train and run entirely on their existing hardware, aiming to automate routine decisions while keeping data within their own systems, the company said in a release.
The company described AT0M as a Made in Bharat model designed for small, repeated business decisions, such as routing requests to the appropriate team, determining whether an alert requires attention and assessing the severity of a reported problem. Unlike AI models that generate free text, AT0M selects responses from options defined by developers. This allows organisations to automate routine cases while referring uncertain or high-stakes decisions to a person for review, the release said. Because it never generates free text, every response stays within the options the developer defined, the release stated. Pienomial said organisations own the model outright, with no per-seat or per-token charges. The model trains and runs within their environment, without requiring data to leave their systems. Most business decisions are small, well-defined, and repeated thousands of times a day. They don't need a giant hosted model. They need something reliable that you own and control. That's what AT0M is, said Sanat Mohanty, CEO of Pienomial. According to the company, AT0M runs as a single executable with no network dependencies and supports computing platforms including Intel Xeon, Apple Metal and NVIDIA CUDA. The company said the model could serve organisations where data must remain on premises, including banks, government agencies, and healthcare and life sciences companies. AT0M belongs to a category the release calls System One AI, which makes structured decisions for software rather than generating text. In company-run testing on the public typed-decisions benchmark, AT0M matched reference answers on 78.9 per cent of 2,000 decisions, the release said, citing a reported score of 74 per cent for TypeSafe's Jev. Pienomial also said AT0M produces reproducible responses and recorded a response time of under 16 milliseconds in a hosted test using an NVIDIA T4 graphics processing unit. (ANI)
|
