Sofia Rahman

Raizen developer profile — @sofia_rahman

ML Engineer at Duolingo

LLM-as-judge reliability. I publish the disagreement, then I fix the prompt.

Location: Atlanta, GA

Ranked #8 on the Raizen verified-developer leaderboard.

Website: https://github.com/sofiarahman

Experience

• ML Engineer at Duolingo

  2021-05-03 – Present

  Automated essay/speech scoring with human-in-the-loop audits.

Education

• Georgia Tech

  M.S., Computer Science

  2018-08-15 – 2020-05-08

Verified skills

• nlp
• django
• python
• pytorch
• prompting

Projects

• Judgejury

  Multi-judge ensemble with item-response theory. Flags items that no judge agrees on.

• Citework

  Forces an LLM grader to quote file:line from a submission before assigning a score.