Vertex AI Clinical Note Classification
Budget: $15 – $25 USD
I want to stand up a new Google Cloud project that automatically classifies clinical notes into the categories I already use internally. The work centres on Vertex AI’s text-classification capabilities, from data preparation through to an endpoint ready for real-time and batch predictions.
Here is how I picture the engagement flowing:
• Spin up and configure the GCP project, set IAM so I can audit all activity and stay HIPAA-ready.
• Import a sample of de-identified notes, split them into train/validate/test sets, and register the dataset inside Vertex AI.
• Train a text-classification model (AutoML or custom if you believe it will perform better), fine-tune until we meet agreed metrics, and export a concise evaluation report.
• Deploy the model to an endpoint, wire up a simple REST or Cloud Function trigger so I can hit it with new notes and receive the category predictions.
• Hand over clean, reproducible code (Terraform or gcloud scripts for infra, Python notebooks or pipelines for training) plus a short run-book so my team can maintain and retrain in the future.
Acceptance will be based on:
1. End-to-end workflow reproducible inside my GCP account.
2. Classification F1 ≥ 0.85 on the held-out test set.
3. Clear documentation covering setup, retraining, and prediction calls.
If this aligns with your Vertex AI and NLP skill set, let’s get started—I’m ready to grant project access and provide the labelled category schema right away.
Here is how I picture the engagement flowing:
• Spin up and configure the GCP project, set IAM so I can audit all activity and stay HIPAA-ready.
• Import a sample of de-identified notes, split them into train/validate/test sets, and register the dataset inside Vertex AI.
• Train a text-classification model (AutoML or custom if you believe it will perform better), fine-tune until we meet agreed metrics, and export a concise evaluation report.
• Deploy the model to an endpoint, wire up a simple REST or Cloud Function trigger so I can hit it with new notes and receive the category predictions.
• Hand over clean, reproducible code (Terraform or gcloud scripts for infra, Python notebooks or pipelines for training) plus a short run-book so my team can maintain and retrain in the future.
Acceptance will be based on:
1. End-to-end workflow reproducible inside my GCP account.
2. Classification F1 ≥ 0.85 on the held-out test set.
3. Clear documentation covering setup, retraining, and prediction calls.
If this aligns with your Vertex AI and NLP skill set, let’s get started—I’m ready to grant project access and provide the labelled category schema right away.