AIRS ML is building the Physical AI layer for industrial machines. We develop edge AI systems that capture high-frequency machine data, process it directly beside the asset and convert it into real-time health insights. Our mission is to make existing machines intelligent, resilient and easier to maintain without requiring continuous cloud connectivity or the replacement of industrial infrastructure. Industrial companies face an enormous and persistent problem: unexpected machine failures. Unplanned downtime costs the world's largest companies an estimated PS1.15 trillion annually, while an idle automotive production line can cost approximately $2.3 million per hour. Beyond the immediate financial loss, downtime disrupts production, increases maintenance costs, affects product quality, creates safety risks and reduces the useful life of expensive equipment. Existing monitoring approaches do not solve this problem at scale. Periodic inspections and low-frequency sensors often detect faults too late. Cloud-based platforms require continuous connectivity and the transfer of large volumes of raw sensor data, increasing bandwidth, storage, latency and cybersecurity requirements. Many machine-learning solutions also depend on years of labelled failure data, which is rarely available for new, bespoke or highly reliable equipment. AIRS ML is based on a simple insight: machines rarely fail without warning--they "whisper" first. Early degradation is often visible in subtle, high-frequency changes in vibration, current, temperature and other physical signals long before a failure becomes obvious. The challenge is not simply collecting more data; it is capturing the right data and processing it at the machine in real time. Our proprietary Edge Processing Systems are installed alongside existing industrial assets and connected to relevant sensors. They capture high-frequency signals, run AI inference locally and convert large raw data streams into compact alerts and actionable machine-health insights. Instead of continuously transmitting raw data to the cloud, AIRS ML sends only the information that matters. This enables low-latency decisions, dramatically reduces data requirements and allows the system to operate in disconnected, air-gapped, remote or highly regulated environments. Our models learn the normal behaviour of each asset using unsupervised and normal-data-only approaches. Customers therefore do not need to wait for breakdowns or assemble extensive historical failure datasets before deployment. Once a normal operating baseline has been established, the system can identify deviations, emerging anomalies and early fault signatures. This supports predictive maintenance, condition monitoring, process optimisation and ultra-low-data-rate digital twins. AIRS ML provides an integrated edge-native platform combining proprietary hardware, embedded AI, sensing, signal processing and industrial deployment workflows. This differentiates us from software-only analytics companies, cloud-heavy predictive-maintenance platforms and sensor providers that leave customers to assemble their own systems. Our repeatable hardware and software architecture can be deployed across different machines while remaining focused on the sensor modalities and asset classes where high-frequency signals provide the greatest value. Our initial market is high-value rotating, electromechanical and production-critical equipment. This includes CNC machines, robotic systems, motors, drives, pumps, compressors, rail assets, agricultural equipment, EV-charging infrastructure and other machines where failure has a significant operational or financial impact. Our primary customers are manufacturers, mobility and rail operators, agricultural and heavy-equipment companies, energy operators and industrial technology providers. The typical buyers are maintenance, reliability, operations, engineering and digital-transformation teams responsible for improving asset availability and reducing maintenance costs. AIRS ML has progressed from laboratory development to validation in real industrial environments. Our systems are installed on three CNC machines at AMRC, where data collection is active. Through Digital Catapult's Digital Twin Adoption Accelerator, we are deploying with AML Sheffield in an industrial programme involving organisations including Artemis, Thales and Short Brothers/Boeing. We are a John Deere Startup Collaborator 2026, with paid-pilot pathways under development, and we are working with Caledonian Sleeper on brake-pad predictive maintenance across four coach types. AIRS ML was also selected as an ABB Startup Challenge finalist for applications involving motors and drive systems. Our technology has demonstrated strong performance across multiple asset classes. Normal-data-only models achieved 96% accuracy on robotic-arm applications and 94.9% accuracy on CNC milling, while our work on motor drives detected both evaluated ABB fault classes with 100% accuracy. These results demonstrate that the same underlying edge AI platform can address different industrial machines without requiring labelled failure data. Our business model combines upfront deployment revenue with recurring software income. Customers initially pay for a pilot that includes Edge Processing Systems, sensors, installation, integration and technical support. Once the pilot demonstrates measurable value, deployments expand from an individual asset to a production line, site and ultimately an equipment fleet. Customers pay a one-time hardware and setup fee for each monitored asset, followed by an annual subscription for monitoring, AI-based fault detection, alerts and operational insights. Our direct go-to-market strategy focuses on paid pilots for high-consequence assets where the cost of downtime creates an urgent and measurable return on investment. We work closely with maintenance and operational teams to define the target asset, baseline performance, success metrics and commercial scale-up pathway. Following validation, we expand within the customer through repeatable installations across similar machines, lines and locations. In parallel, we are developing partnerships with industrial OEMs, automation companies and system integrators. These organisations can embed, integrate or resell AIRS ML across their installed equipment bases, accelerating distribution and enabling fleet-scale deployment. This channel creates opportunities for hardware sales, recurring software subscriptions, licensing and embedded intelligence within new industrial equipment. The market opportunity sits at the intersection of predictive maintenance, edge computing and Industrial IoT. The predictive-maintenance and edge-computing markets are projected to reach PS48.9 billion by 2030 and PS109.4 billion by 2032. The wider Industrial IoT manufacturing market is projected to grow from approximately $97 billion in 2023 to nearly $674 billion by 2032. AIRS ML is positioned to capture this opportunity by turning underused machine data into real-time operational intelligence. Our team combines advanced AI, semiconductor and embedded hardware engineering, signal processing and real-world industrial deployment experience. CEO Dr Prateek Tripathi is a former Senior ASIC Engineer at Qualcomm with experience in edge hardware, industrial programmes and Innovate UK projects. CTO Dr Soumya Gupta holds a PhD from the University of Oxford and brings expertise in artificial intelligence, computer vision and commercial technology development from Boston Scientific. Our embedded systems capabilities include experience in PCB design, signal processing and edge deployment developed at organisations including NXP and Beckman Coulter. AIRS ML is backed by Techstars and has been supported by leading UK and European industrial innovation organisations and programmes, including Digital Catapult, the UK Digital Twin Centre, HVM Catapult, Innovate UK, DRIVE TLV and EIT Urban Mobility. Our long-term vision is to provide the intelligence layer for the machines that already power the physical world. By processing data where it is generated, AIRS ML enables industrial companies to detect problems earlier, act faster and scale AI across thousands of assets without overwhelming their networks, cloud infrastructure or engineering teams. We aim to make real-time machine intelligence a standard capability of every critical industrial asset.