Bob Killer on Snowflake

Every company runs on someone's spreadsheet: Bob's 2013 workbook, full of formulas and macros nobody else understands, undocumented, untested, and one bad paste away from a wrong number in a board deck.

Bob Killer turns that workbook into real code you own. Upload it in Snowsight, and Snowflake pulls it apart:

  • every cell, formula, lookup and macro;
  • the dependencies between them;
  • hardcoded numbers, promoted to named parameters.

You get back a dbt project that runs in Snowflake, with plain-English docs and parity tests built in.

What sets it apart is proof. Every output is checked against Excel's own results with an exact comparison: the same number, bit for bit, no tolerances. Anything it can't convert is flagged in a coverage report, never quietly approximated. Bob can still change the numbers, and you get a tested pipeline in your warehouse instead of a file on someone's desktop.

Status: the upload, landing and run-tracking platform is merged, along with the extract and formula-graph stages. The Excel runtime and the dbt generator are still being built. The full workbook-to-dbt run with exact parity hasn't been proven yet.

Built With

  • actions
  • claude
  • cli
  • container
  • cortex
  • dbt
  • fastapi
  • github
  • hypothesis
  • jinja
  • odcs
  • openpyxl
  • pydantic
  • pytest
  • python
  • snowflake
  • snowpark
  • sql
  • sqlite
  • streamlit
  • uv
Share this project:

Updates

Submission history