The Problem We Are Solving: African manufacturing faces a specific set of constraints that generic cloud AI does not address
Why this challenge Matters for Coo-Cah? Unstable internet connectivity: A factory cannot halt if the cloud is unreachable High energy cost and load-shedding AI must help optimise energy use, not just consume it.
Local data control requirements: African data protection regulations (NDPA, Rwanda Data Protection Bill) require in-country residency Multi-site replication at low cost: The same factory AI stack must replicate to 17 factories without 17 separate cloud budgets Why Arm? Arm-powered edge hardware (Arm64 servers, embedded SoCs) is the natural fit for Coo-Cah's factory edge nodes because:
-Energy efficiency: Lower watt-per-inference than x86, critical for solar-powered factories -Ecosystem maturity: NVIDIA Jetson (Arm), AWS Graviton (Arm64), and single-board compute all run on Arm -Cost-performance: Arm64 edge nodes deliver sufficient inference throughput at a fraction of the GPU cloud cost -AfricaArm alignment: Arm-based hardware is accessible and affordable across African markets.
Track Positioning
Our Claim Physical AI (primary): Edge AI running on factory hardware, connected to real sensors, AMRs, and MES systems — real-world physical systems Cloud AI (secondary): Edge syncs summaries to a Rwanda cloud hub; fleet learning and cross-factory AI training happen there Mobile AI: Not in scope for this submission
What We Are Building
A focused demo of the Coo-Cah Edge Twin operating on Arm64 hardware.
-Simulated IoT sensor telemetry (vibration, temperature, current) for an SMT line machine in the Personal Electronics pilot factory (Sagamu, Nigeria) -Local predictive maintenance model running on the Arm edge node — anomaly detection + failure risk score -Energy-aware scheduling signal — edge system recommends whether to run the next production batch based on current solar/BESS state -Offline continuity proof — demo shows the system working with no cloud connection -Cloud sync — summaries and model feedback sent to Rwanda hub when connectivity is available.
Scope Constraints To stay buildable within the hackathon window:
Dimension Constraint
Factories: 1 Personal Electronics, Sagamu Machine types: 2 SMT pick-and-place + reflow oven AI models: 1 Predictive maintenance (anomaly detection + RUL estimate) Dashboard: 1 Grafana local + cloud sync view
KPI story: 4 metrics
KPI Target -Edge inference latency: < 50 ms per sensor window on Arm64. -Cloud bandwidth reduction: ≥ 80% vs. raw stream upload (edge aggregation). -Fault prediction lead time: ≥ 48 hours ahead of failure. -Energy recommendation accuracy: ≥ 85% correct batch-start decisions vs. manual baseline .
Log in or sign up for Devpost to join the conversation.