20 January — MEDTRAC™ has developed a proprietary knowledge-retention platform, the MEDTRAC™ Knowledge Surface, designed to train the company’s large language model and codify operational and analytical rules across its projects. The platform addresses the persistent “Where did I put that information?” problem by routing all relevant data to a single, structured location and enabling high-speed search with typical retrieval times of 0.2–0.5 seconds, combined with large-language-model–based discovery. This approach removes the need to continually retrain models on locally stored data.

The MEDTRAC™ Knowledge Surface supports rapid publication of website content, ingestion and syndication of both internal and external knowledge sources, and fast retrieval via a high-performance search engine. It is fully integrated into the company’s website to enable rapid publishing, SEO optimisation, and content syndication, and is intended to replace the company’s use of OneNote, traditional file storage, and standalone chat systems.

The platform enables secure ingestion and rapid search across both external and internal knowledge sources, including patient and product data, while remaining accessible to internal models and cloud-based large language models such as OpenAI, Anthropic, Grok, and Llama.

The system is powered by the company’s local LLM service hosted in Mellieħa, which will be upgraded from a 7-billion-parameter model to a 20-billion-parameter model. The local LLM will be used for on-premise processing of low- to medium-complexity workloads, reducing typical LLM response times from approximately 10–30 seconds to 1–5 seconds.