Moii AI

Background

Moii AI, an innovative Vision AI startup, needed to accelerate the training of its real-time object detection models. Their existing manual video annotation process was slow, labor-intensive, and unable to keep up with rapidly growing video datasets. High-quality, consistent labels were critical for model accuracy, and the team required a scalable, cost-efficient pipeline on Google Cloud to support ongoing development and growth.

Challenges

Key challenges included:

• Slow Manual Annotation: Video labeling was time-consuming, delaying model training cycles.
• Rapidly Scaling Video Volume: The existing workflow could not keep pace with the growing dataset.
• Label Consistency & Accuracy: High-quality annotations were essential for reliable object detection.
• Cost-Efficient, Reliable Infrastructure: A fully managed, scalable GCP environment was needed under tight timelines.

Solutions Delivered

Zazmic designed and deployed a secure, end-to-end AI-powered video annotation pipeline on Google Cloud:

• Automated Video Ingestion & Management: Video retrieval and annotation workflow automated via Cloud Scheduler, Cloud Run, and Cloud Functions.
• AI-Powered Labeling: Gemini AI used for Visual Question Answering (VQA) and pseudo-labeling to generate annotations automatically.
• Serverless Orchestration: Cloud Functions trigger detection jobs on auto-stopping VMs, ensuring compute runs only when needed.
• Full GCP Environment Setup: BigQuery, Cloud Functions, Cloud Run, logging, and monitoring implemented for a scalable, maintainable, and production-ready system.

Outcomes

The new pipeline transformed Moii AI’s data operations:

• Faster Annotation: Automation drastically reduced labeling time.
• Scalable Processing: BigQuery enables high-speed handling of large and growing video datasets.
• Improved Accuracy: Gemini-assisted labeling enhances label quality, boosting model performance.
• Automated Orchestration: Cloud Functions and serverless workflows handle ingestion, retrieval, and processing automatically.

Conclusion

Zazmic built a scalable, AI-powered video annotation pipeline that transforms Moii AI’s data operations. By leveraging Gemini and Google Cloud’s serverless capabilities, Moii AI can now process large video datasets efficiently, generate high-quality labels for real-time Vision AI models, and prepare for future growth—from hundreds to thousands of cameras—while keeping costs optimized.