Keynote: Empowering Accessibility Through Kubernetes: The Future of Real-Time Sign Language Interpretation
Talkat KubeCon + CloudNativeCon Europe 2025 · on Wednesday, 2025-04-02· at 10:01· Level 0 | ICC Auditorium·
summary
Demonstrates a real-time sign‑to‑speech system deployed on Kubernetes that uses CubeFlow pipelines, computer‑vision landmark extraction via MediaPipe/MM Labs, and kserve for optimized inference while consulting LLMs to resolve context (e.g., Austin vs Albany). Shows X/Y/Z hand landmarks, shift-and-scale normalization, multiple camera angles to fix P vs Q errors, and a live demo that successfully outputs spoken phrases. Argues Kubernetes handles video+AI scaling and GPU/CPU placement but stresses data collection and view diversity as the current bottlenecks.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:03Challenges in sign recognition
- 2:52Kubeflow for ML workflows
- 4:04AI processing architecture
- 7:08Kubernetes orchestration
- 9:06Demo and data optimization
- 11:36Authenticity in training
- 13:34Roadmap and conclusion
related talks
27mSign Language Recognition Improvements Since the London Keynote
16mKeynote: An Open-Source Stack for AI Compute: Kubernetes + Ray + PyTorch + vLLM
32mKeynote: Powering Generative AI with Kubernetes
23mKeynote
40mKeynote: Open-source Multimodal AI in the Wild
18mKeynote: Adapting Kubernetes for Machine Learning Workflows
15mKeynote: Machine Learning on Kubernetes Made Easy With Kubeflow
21mKeynote: A Tale of Using Kubeflow to Make the Electricity Smarter in China