AI Atlas - Daily updates on real-world AI deployments worldwide

Citymind's AI chatbot on the Žilina city website received 8,603 questions from approximately 1,850 citizens over a six-month pilot from 1 July 2025 to 1 January 2026, with 47.3 percent of conversations occurring outside business hours and only 3.1 percent of inquiries left unanswered, per research by the Faculty of Public Administration of Pavol Jozef Šafárik University in Košice. The strongest topic was the building authority at 98.8 percent resolution, while cultural information scored lowest at 44.42 percent, exposing a website content gap. The deployment, led by Mayor Peter Fiabané and researched by Milan Douša, PhD, is positioned as a model for other Slovak cities seeking data-driven AI deployment.

Citymind's AI chatbot on the Mokrá-Horákov municipal website has handled nearly 13,000 inquiries across more than 6,900 conversations since launch, correctly resolving 90.9 percent of relevant inquiries and saving the office over 670 hours. Mokrá-Horákov, a small village near Brno, is likely the first municipality in the Czech Republic to deploy an AI chatbot for citizens, with the 1 May 2024 launch covered by Czech Radio Radiožurnál. The deployment is positioned as a model for small Czech villages seeking to digitize citizen services without large IT investment.

Citymind's AI chatbot at the Mendel University in Brno (MENDELU) Faculty of Economics and Management has handled over 65,200 questions with 93.8 percent accuracy, saving the university more than 1,713 hours, with more than 50 percent of conversations occurring outside business hours. The deployment started as a pilot at the Faculty of Economics and Management (PEF) and has since expanded to dormitories, the canteen, the IT department, and the sports centre. MENDELU was Citymind's birthplace as a 2023 university course project, and the first MENDELU Boost student competition was won by the Citymind project.

Citymind's AI chatbot on the O2 arena website in Prague correctly handled more than 49,400 visitor inquiries during the review period, with a 93 percent resolution rate and more than half of all conversations occurring outside working hours. The deployment saved over 2,345 hours of manual processing, with the top topics being tickets, event start times and entrances, parking and transport, pre-visit operational information, and event-specific orientation. The case demonstrates that AI chatbots work in user-impression-critical environments like event venues, not just government and education.

Citymind's AI chatbot deployed in Prague 6, a district of the Czech capital, identified identity cards as its strongest single topic at 16.1 percent of all queries with 75.6 percent answer accuracy, while receiving an average user rating of 4.05 out of 5. The deployment was funded through the EU Commission's CommuniCity project, which pre-selected quality AI vendors for European cities. The case study highlights the chatbot's role as both a citizen service and a feedback loop into website and communication improvements via Citymind Manager's unanswered-questions analytics.

KingMed Diagnostics and the First Affiliated Hospital of Guangzhou Medical University jointly released DeepGEM 2.0, a multimodal AI model that predicts lung cancer driver gene mutations from routine digital pathology slides in under one minute, covering all gene targets corresponding to domestically approved targeted therapies. The model is being lightweight-deployed across 500+ Chinese medical institutions via KingMed's KMDP digital pathology platform, with development timelines of 12-16 weeks for new gene additions when new drugs receive regulatory approval.

Oregon Health Authority's Office of Information Services (OIS) deployed IBM Instana Observability on its eXPRS payment application and reduced mean time to detect by 50% and mean time to resolve by 30%, while serving 30,000+ individuals with disabilities.

JPMorgan Kinexys and BNY jointly explored federated learning for cross-institution anomaly detection in a permissioned PoC using synthetic data, with the aggregated model converging by the fourth aggregation round and detecting anomalies beyond either participant's local model.

FNBO deployed Pindrop Fraud Assist — a GenAI add-on to Pindrop Protect — at its fraud investigation operations from Aug–Dec 2025, achieving 35–40% faster case reviews, 50% higher decision accuracy, and 42% more cases handled without adding staff.

VA's National Healthcare Review (Report 26-00182-140, June 11, 2026) found VHA operates two general-purpose generative AI chat tools — VA GPT and Microsoft 365 Copilot Chat — across clinical workflows, drawing 15,832 active users across two internal Microsoft Teams communities between October 9, 2025 and January 6, 2026. The OIG issued three recommendations to classify these tools as high-impact and integrate AI risk monitoring into patient safety programs by April 2027.

Rakuten Mobile and Rakuten Symphony deployed an in-house RAN Intelligent Controller (RIC) across Rakuten's nationwide 4G and 5G Open RAN network in Japan — one of the world's first nationwide commercial RIC deployments — with ML-powered rApps that predict subscriber traffic and dynamically adjust base station utilization for a targeted 15–20% network energy reduction.

Stanford Health Care rolled out Microsoft DAX Copilot, an ambient AI scribe integrated with Epic, in its emergency department in stages from February 2025. A retrospective cohort of 10,344 encounters and 100 attending physicians found scribe use associated with a 72.6-second reduction in on-shift documentation time per encounter, about 28% below the modelled baseline of 260 seconds.

Rabobank's Digital Platform engineering team has been experimenting with AI agents and multi-agent architectures on Azure AI Foundry and the Microsoft Agent Framework to evaluate production readiness. The team documents recurring challenges — supervisor-agent dilution, tool-call failures, non-deterministic orchestration — and shares lessons on prompt design, model selection and continuous evaluation.

ING and Worldline completed Europe's first live agentic payment in production at the Money20/20 Europe event on 2 June 2026, with an AI-powered merchant agent initiating a purchase from an ING cardholder using Mastercard rails and Worldline's pan-European processing infrastructure.

Decathlon runs Zowie AI customer-service agent in production, autonomously resolving 98.47% of support conversations while driving +20% support-originated revenue and +8% conversion, with deflection rates growing from 30% to 50% year-on-year and email integration handling 55% of Polish customer contact volume.

ThinkDeep AI built DeepBrain, a sovereign multi-agent platform using NVIDIA AI Enterprise, NIM microservices, NeMo Retriever, NeMo Guardrails, and DGX H100/H200 hardware, cutting French government document processing from 2 days to 2 minutes and saving €2M at the Ministry of Economy and Finance at 10,000-employee scale.

Thames Valley Police and Hampshire & Isle of Wight Constabulary deployed Salesforce Agentforce to handle 200+ non-emergency conversations daily with 75% autonomous resolution, plus a MuleSoft + Agentforce Experience citizen portal that cut call volumes by 10% across 4 million residents.

German commercial and industrial insurance broker MRH Trowe gave ~400 employees secure self-service access to AI agents built on Amazon Bedrock AgentCore + Strands Agents + LibreChat, hitting $14 per seat per month with a clear path to ~40% infrastructure-cost reduction while keeping client data inside the AWS Europe (Frankfurt) region.

A US six-hospital, 1,000-bed nonprofit academic health system deployed Harrison.ai Comprehensive Care (AI OHCP triage on head CT) and captured more than $100K in Medicare NTAP reimbursement across its network in roughly four months.

Austrian trade union GPA deployed checkts.gpa.at, an AI chatbot for labour law questions, with 88% of users opening the cited paragraph and 12% clicking through to the federal legal information system.