The Institut du Cancer de Montpellier’s deployment of Miroki, an AI-enabled companion robot running on Microsoft Azure and Azure OpenAI Service, is more than a feel-good healthcare pilot. It is a carefully architected solution that tackles one of pediatric oncology’s most difficult challenges: keeping children still during radiation therapy without sedation.

The Problem: Motion During Pediatric Radiation

Radiation therapy demands absolute stillness. For adults, a few minutes of lying motionless is manageable. For children, especially those under seven, it’s nearly impossible without general anesthesia. Repeated sedation carries risks: respiratory complications, delayed development, and increased procedure times. The Institut du Cancer de Montpellier, part of the University Hospital of Montpellier, treats roughly 50 children per year requiring radiation. Many of these children need daily sessions for several weeks. Reducing anesthesia exposure became a clinical priority.

Enter Miroki: An AI Companion, Not a Toy

Miroki is not a generic chatbot on wheels. Developed by French startup Miroki, this companion robot integrates with Azure OpenAI Service to deliver real-time, emotionally responsive interactions. The robot uses a large language model to generate age-appropriate conversation, answer questions, and guide breathing exercises. It runs on Azure infrastructure, ensuring low latency and HIPAA-compliant data handling.

The robot’s design is intentionally non-threatening. It resembles a friendly animal with large eyes and a soft voice. During a procedure, Miroki sits on a table next to the child, maintaining eye contact and adapting its dialogue based on the child’s emotional state. If the child becomes anxious, Miroki shifts to calming stories or breathing prompts. If the child is curious, it explains the machine noises in simple terms.

Technical Architecture: Azure OpenAI in Healthcare

Under the hood, Miroki leverages Azure OpenAI’s GPT-4 model, fine-tuned for pediatric communication. The system ingests real-time audio from the child, processes it through Azure Speech Services for transcription, and routes the text to the LLM. The response is then synthesized into speech via Azure Text-to-Speech, with emotional tone modulation. The entire loop completes in under two seconds.

Data privacy was paramount. The hospital worked with Microsoft’s healthcare compliance team to ensure the solution met European GDPR and French health data regulations. Patient data remains within the hospital’s Azure tenant, and no audio or transcripts are stored outside the clinical environment. The robot does not connect to the internet; all processing occurs within the hospital’s secure network.

Real-World Results: Reduced Sedation, Increased Compliance

The pilot program, launched in late 2024, included 12 children aged 3 to 7. The results were striking: 8 of 12 children completed their radiation sessions without any sedation. The remaining 4 required reduced doses. Parents reported significantly less pre-procedure anxiety. One mother described her daughter “talking to Miroki like a friend” during the 15-minute sessions.

Dr. Sophie Laurent, head of pediatric radiation oncology at the institute, noted: “Before Miroki, we would spend 30 minutes preparing a child for anesthesia. Now, many walk into the room holding Miroki’s hand. The robot absorbs their fear.”

Challenges and Limitations

The pilot also revealed issues. Some children were initially wary of the robot, requiring a familiarization session the day before. The LLM occasionally generated responses that were too complex for very young children, prompting the team to add age tiering to the prompt. Azure OpenAI’s safety filters also flagged certain innocent phrases (like “I’m scared of the big machine”) as potentially harmful, requiring manual override by the clinical team.

Latency, while generally under two seconds, sometimes spiked during network congestion, causing awkward pauses. The team is exploring Azure’s edge computing options to further reduce lag.

Broader Implications for Healthcare AI

Miroki’s success at Montpellier is part of a larger trend. Microsoft has been aggressively positioning Azure as the platform for healthcare AI, with dedicated offerings like Azure Health Data Services and the Microsoft Cloud for Healthcare. Companion robots, however, represent a new frontier: deploying LLMs in real-time, patient-facing roles where emotional intelligence is as critical as factual accuracy.

The Montpellier team plans to expand the pilot to more children and eventually to other stressful procedures like MRI scans. They are also developing a version for autistic children, who often struggle with unfamiliar clinical environments.

The Bottom Line

Miroki demonstrates that AI companions can move beyond novelty into genuine clinical utility. By running on Azure OpenAI, the robot combines state-of-the-art language understanding with enterprise-grade security. For pediatric patients, it offers a way to reclaim a sense of control in an otherwise frightening experience. For healthcare systems, it reduces the burden of anesthesia and speeds up procedure times. The technology is not perfect, but the trajectory is clear: AI companions will become standard tools in pediatric care, and Azure is poised to be the platform that powers them.