PandasAI, Data Formulator and Claude workflow overview

The speaker surveys how AI tools fit into data science work, starting with exploratory analysis. PandasAI accepts natural-language questions about a dataframe and can return charts, tables or other answers. In the example, an OpenAI model receives the question, column names, sample rows and metadata, then generates Python that PandasAI checks and executes locally. This still sends some data context to the model. The speaker recommends using Pandas directly for more complex operations such as joins and correlation matrices.

Data Formulator, from Microsoft Research, shows the steps behind data transformations and visualizations. The speaker describes both tools as open source and says Data Formulator can be installed locally. These examples explain local execution, but do not establish an offline workflow.

Other examples include using Claude to generate a 5,000-row e-commerce dataset and trying foundation models for forecasting. The speaker names Amazon's Chronos and describes obtaining a forecast without a separate training loop in the example.

An MCP example connects Claude Desktop to PostgreSQL for questions about customer revenue. The speaker cautions that employer security policies may prohibit Claude Desktop or Claude Code. The final discussion covers building usable AI systems, Docker containerization, deployment and monitoring, alongside business judgment and communication.