Researchers have published multiple papers on arXiv introducing specialized AI systems designed to address specific challenges in safety-critical and resource-constrained domains.
Medication Safety for Older Adults
According to arxiv.org, a system called ATLAS introduces “a coupled graph—policy distillation framework for patient-adaptive medication safety” targeting older adults with multimorbidity. The framework structures guideline evidence as a medication-safety graph and uses targeted questions to create patient-specific medication conflict graphs. On a European non-interactive multimorbidity benchmark, ATLAS “exceeds the strongest proprietary LLM baseline by 53.73 points in Strict Success Rate and 14.63 points in overall safety reasoning score,” with no unsafe recommendations under automated evaluation.
Mass Gathering Assistant
A separate arxiv.org paper describes KumbhDoot, an assistant designed for mass religious gatherings like the Kumbh Mela. The system “operates on a foundational design principle that prioritizes semantic similarity over starting with an LLM,” invoking generative models only when similarity-based retrieval proves insufficient. According to the researchers, this approach is “not merely cheaper but architecturally more appropriate” for bounded, high-stakes, low-connectivity public-service domains.
Additional Systems
Arxiv.org also published research on SodaMem, an evidence-grounded temporal graph memory system that achieved 92.8% accuracy on LongMemEval-S, and LayoutDSL, which uses domain-specific language for interior layout design reasoning rather than direct coordinate prediction.