Work

11 projects, numbered straight through.

Every number is tagged with how it was measured. Projects without a case study link to their code, or say why there is none yet.

Showing all 11 projects

  1. 01 Product Guidance Reconciliation Engine Did the company hit the revenue guidance it gave last quarter? measured 7 of 188 sent to the LLM Live · 97.9% on hand labels
  2. 02 Open source faultline Why did the dbt pipeline break, and where is the evidence? measured 89.6% top-1, 0 wrong (dev) In progress · held-out pending
  3. 03 Product Werkbank Which workshop should take this furniture order? measured 95.8% on auto-routed orders Live demo
  4. 04 Research FinBERT fusion When should a small model hand a sentence to an LLM? measured 0.878 accuracy (FinBERT alone 0.844) Case study coming
  5. 05 Tool Sustainability commitment classifier Which climate commitments in a sustainability report come with a number? measured 12.3% of 1,562 commitments quantified Done · rule-based
  6. 06 Research DAX 40 capital structure What drives leverage in large German companies? measured 127 firm-years, 32 firms · graded 1.3 Paper · code being published
  7. 07 Research ESG and bond spreads Do ESG rating changes move corporate bond spreads? measured No significant effect (p 0.53–0.97) Done · honest null result
  8. 08 Product Dem Fragen What does Indian law say about my problem, and when do I need a lawyer? pending — real eval not run yet Live · in progress
  9. 09 Tool GradeLens What does each remaining FOM exam need for my target grade? measured v0.4 unit and end-to-end tests Chrome extension
  10. 10 Tool profile-hud Can a GitHub profile show real project data instead of badges? measured 156+ tests, CI green GitHub Action · MIT
  11. 11 Open source This website Can a portfolio prove its claims instead of asking you to trust them? measured 100 Lighthouse accessibility, home page Live · case study