I am an FDE and in all my clients' projects I have encountered the same problem: real world data preparation is the bottleneck. Company data doesn't only come with the usual Kaggle problems, it also has a wide range of other problems: corrupted migrations, multiple versions of taxonomies, duplicate columns and many more.
Each problem requires a decision and portia helps accelerate all those and keep everything in a long term memory making work sessions persist from one chat to another. Today portia works with Claude Code, Codex or a local model (llama.cpp or ollama). The data can be a Postgres database, CSV files or hosted in the cloud (Snowflake or BQ).
Happy to answer any questions!