The team of the NeoLens project has released an update to its AI assistant for the diagnostics and repair of military equipment on an Android basis, which allows the application to work without an internet connection.
Previously, military mechanics and technical specialists could receive repair advice only when communication was available. The new version solves the problem of working in field and combat conditions, where an internet connection is limited or absent.
According to NeoLens CTO Ostap Korkuna, adapting AI to work without a network required solving two main problems: high hardware requirements and the lower quality of local language models.
"The first versions required the newest phones with built-in graphics accelerators (the latest iPhone/iPad or a special gaming Android), which made the solution completely inapplicable in real combat conditions in Ukraine. In addition, there was a huge difference in the quality of the assistant's responses based on models that can run on a phone versus the online version that uses advanced large models," noted Ostap Korkuna.
He explained that small language models are developing more slowly than frontier solutions from OpenAI or Google, since large laboratories invest resources in them on a residual basis. To achieve the necessary accuracy on ordinary smartphones, the developers combined optimized AI models, a specially processed knowledge base, and experience from prior use of the online version.
Before the release, the offline version was tested by a group of military mechanics. According to their feedback, the assistant works with a slight delay and provides structured instructions for checking components, particularly during the diagnosis of engine malfunctions.
NeoLens launched an AI assistant for the diagnostics and repair of Humvee, MaxxPro, and M113 armored vehicles last year. The company reported that the assistant is used in dozens of Ukrainian repair battalions along the entire front line and is gradually being scaled to other types of equipment, including robotic platforms and artillery. The system was most actively used on Humvee platforms.
