- Describe your company in one paragraph.
Plaixus is a Greek deep-tech SME, founded in 2022, specialising in industrial-grade machine learning, MLOps and edge-AI applications, with a particular focus on computer vision and data analytics for industrial and agricultural settings. The company operates across the full ML lifecycle, from data ingestion and model training to edge deployment and monitoring, turning research-grade concepts into field-deployable products. Plaixus has already delivered several EU-funded R&D projects, including ELDER (Edge-AI for Lameness DEtection in Ruminants), a cascade-funded experiment under the AGRARIAN programme, which established our foundational expertise in applying computer vision and edge-AI for livestock monitoring. Our approach is to act as an intelligence layer on top of existing OEM equipment.
- What challenge are you addressing under the O-CEI Horizon’s first Open Call, and how is your proposal relevant to the challenge?
SHEPHERD-AI addresses Challenge CH5.5 under Pilot-5, the Halloumi value chain in Cyprus, where large-scale dairy-cow farms struggle to detect health issues at a sub-clinical stage. In these high-demand farms, cows showing early signs of illness often go unnoticed, and by the time clinical symptoms appear, interventions are limited, costly or ineffective. This is particularly relevant for diseases such as Foot-and-Mouth Disease (FMD), one of the most economically devastating livestock diseases worldwide, where fever is an early indicator, and for heat stress, which reduces milk yield and inflates energy and water consumption per litre produced. Our solution installs thermal/infrared cameras inside milking parlours to measure individual cow body temperature, in a non-invasive way, during routine operations. Deep learning models detect the head, eyes or other body parts to extract accurate temperature readings, which are then fused with parlour signals (milk yield, milking time) and environmental data (ambient temperature, humidity, wind speed) through predictive analytics. Alerts are generated at a sub-clinical stage, enabling low-cost, timely interventions. The proposal is relevant to the O-CEI Open Call on several fronts. The architecture is edge-first: raw video remains on-farm and only compact per-animal metadata is transmitted, addressing digital energy and privacy concerns. All telemetry is published as JSON-LD aligned with SAREF4Agri, ensuring semantic interoperability with the CEI-InOE ecosystem. The system will be listed on the O-CEI Marketplace, turning a pilot into a repeatable offering across regions. Starting from TRL 5, SHEPHERD-AI is expected to reach TRL 7 within the seven-month execution window, through a three-work-package structure covering design, integration and exploitation.
- What is the expected impact of your proposal?
SHEPHERD-AI’s expected impact unfolds across three layers: farm-level productivity and welfare, energy efficiency, and scalability within the O-CEI ecosystem. At the farm level, the system is expected to generate sub-clinical alerts at least seven days earlier than parlour-only deviation detection, with the predictive pipeline achieving AUROC ≥85%, precision ≥0.70 and recall ≥80% in real-world conditions at Pilot-5. The alerting pipeline will deliver notifications within five minutes of detecting elevated temperatures, with dashboard latency of ≤2 seconds and availability ≥99.5%. Over the longer term, we anticipate a milk yield increase of ≥5% and a reduction in veterinary costs of ≥15%, driven by timely, low-cost interventions rather than escalating clinical episodes. In the case of FMD, early detection is particularly consequential, since outbreaks typically lead to large-scale culling and the loss of all resources invested in affected animals. On energy and sustainability, earlier detection reduces the duration of energy-intensive milking sessions, lowers water demand for cleaning and sanitation cycles around sick animals, and enables targeted activation of cooling systems only when heat stress is actually detected. These outcomes align with Pilot-5’s Scope-3 and Green Deal objectives and will be quantified through parlour telemetry and routine logs. At the ecosystem level, all outputs are published as JSON-LD/SAREF4Agri (100% schema compatibility, zero manual wrangling upstream), ensuring interoperability with the O-CEI platform. The system will be listed on the O-CEI Marketplace, with a deployment pathway targeting ≤2 days from installation to first alerts on a new farm. This positions SHEPHERD-AI for scaling beyond Cyprus, along an EU roadmap extending to Greece and, subsequently, Italy and Romania, in partnership with cooperatives and equipment OEMs.
