Thanks to everyone who joined this week’s Devpost webinar and office hours. We had a great session led by Rick Jacobs, who demonstrated how to build and deploy a tourism recommendation system using BERT in AI Studio. The discussion covered key features of the platform, tips for working without a GPU, and clarified project submission requirements. Below is a summary of what was covered and shared during the session.

Live Demo: Tourism Recommendation System in AI Studio
Presented by Rick Jacobs, the session featured a walkthrough of building a semantic search recommendation system using a BERT model from the NVIDIA NGC catalog. The demo included:

• Generating embeddings from a corpus using a pre-trained BERT model

• Deploying the model with MLflow for tracking

• Serving the system locally with Swagger UI for testing

Key Features of AI Studio Highlighted

• Preconfigured environments and containerized workflows

• Support for custom models and datasets

• Built-in deployment tools including MLflow

• GitHub integration for code management and collaboration

Q&A Discussion Points

• You can bring your own models into AI Studio

• Mac support is not available yet but under consideration

• NVIDIA GPUs are not required, but performance may vary without one

• CPU-only optimization tips were shared (quantization, optimized runtimes)

• Computer vision projects are welcomed and judged equally

• Project submissions must include GitHub code, a demo video, and clear documentation

• Submissions must be able to run within AI Studio for evaluation

Reminders

• Project templates and starter kits are available in AI Studio

• Weekly office hours are available for questions and feedback