[
  {
    "id": 0,
    "start_sec": 0.0,
    "end_sec": 7.0,
    "text": "Welcome to this short introduction to MLOps, the practice of taking machine learning models to production.",
    "language": "en"
  },
  {
    "id": 1,
    "start_sec": 7.0,
    "end_sec": 16.0,
    "text": "In this overview you'll see the full pipeline on the slides: data, training, serving, and monitoring.",
    "language": "en"
  },
  {
    "id": 2,
    "start_sec": 16.0,
    "end_sec": 25.0,
    "text": "Everything starts with a good dataset. We collect, clean, and label our training data.",
    "language": "en"
  },
  {
    "id": 3,
    "start_sec": 25.0,
    "end_sec": 35.0,
    "text": "On screen you can see the dataset summary, with the number of samples and the class balance.",
    "language": "en"
  },
  {
    "id": 4,
    "start_sec": 35.0,
    "end_sec": 45.0,
    "text": "Next we move on to training. We fit a simple model and track its accuracy on a validation set.",
    "language": "en"
  },
  {
    "id": 5,
    "start_sec": 45.0,
    "end_sec": 55.0,
    "text": "Here is the training curve. Notice how the validation loss flattens after a few epochs.",
    "language": "en"
  },
  {
    "id": 6,
    "start_sec": 55.0,
    "end_sec": 65.0,
    "text": "Once training is done, we save the model and wrap it in an inference function.",
    "language": "en"
  },
  {
    "id": 7,
    "start_sec": 65.0,
    "end_sec": 75.0,
    "text": "Now let's talk about serving. We expose the model through a REST API so other services can call it.",
    "language": "en"
  },
  {
    "id": 8,
    "start_sec": 75.0,
    "end_sec": 85.0,
    "text": "The API has a predict endpoint that takes a JSON request and returns the model's prediction.",
    "language": "en"
  },
  {
    "id": 9,
    "start_sec": 85.0,
    "end_sec": 95.0,
    "text": "How do we ship this to production reliably? We package everything into a Docker container.",
    "language": "en"
  },
  {
    "id": 10,
    "start_sec": 95.0,
    "end_sec": 105.0,
    "text": "After deployment, monitoring becomes critical. We track latency, errors, and prediction quality.",
    "language": "en"
  },
  {
    "id": 11,
    "start_sec": 105.0,
    "end_sec": 115.0,
    "text": "Let me show you the metrics dashboard, with p50 and p95 latency for each endpoint.",
    "language": "en"
  },
  {
    "id": 12,
    "start_sec": 115.0,
    "end_sec": 125.0,
    "text": "We also add a health check endpoint so the platform can restart the service if it fails.",
    "language": "en"
  },
  {
    "id": 13,
    "start_sec": 125.0,
    "end_sec": 135.0,
    "text": "A quick question: what happens when the data distribution changes over time? That is called drift.",
    "language": "en"
  },
  {
    "id": 14,
    "start_sec": 135.0,
    "end_sec": 145.0,
    "text": "To summarize, MLOps connects data, training, serving, deployment, and monitoring into one pipeline.",
    "language": "en"
  },
  {
    "id": 15,
    "start_sec": 145.0,
    "end_sec": 150.0,
    "text": "Thanks for watching this overview. In the next video we'll build the API step by step.",
    "language": "en"
  }
]