Miju Labs

← All people · Snorkel AI

P

Paroma Varma

Co-founder & Head of Research · Snorkel AI

Snorkel AI co-founder who leads research; her 2025-26 output is squarely on how expert data and rubrics are evaluated (rubric failure modes, automated benchmark design, RLVR in low-data regimes). She is the research owner behind Snorkel's claim that rubric quality is make-or-break.

medium confidenceAt Snorkel AI since 2019Updated 2026-09-19

Background

Stanford PhD in Electrical Engineering from Chris Ré's group; her research aimed at 'democratizing ML for domain experts' without large datasets, with applications in medical imaging and autonomous driving.

What they run now

Research on evaluation of expert data: RIFT (rubric failure mode taxonomy, ICLR 2026 workshop), 'Learning from Less' on RLVR with limited data/compute (MLSys 2026), and BeTaL automated benchmark design (Oct 2025).

Career

  1. 2019 – presentCo-founder & Head of Research, Snorkel AI
  2. Stanford University, PhD, Electrical Engineering

On the record

Rubrics · 2026-04

Rubric-based LLM evaluation has systematic failure modes that need automated diagnostics (RIFT paper). source

Programmatic labeling

Domain experts should encode knowledge as programmatic signals rather than hand-label at scale. source

Interviews & talks

Connections

Who they amplify

Accounts whose posts Paroma has reposted recently: Alexander Ratner, Henry Kiss Ehrenberg.

Why it matters here

Her rubric-failure research is directly transferable to taste data, where rubrics are the hardest part (subjective criteria, inter-rater disagreement). Reading RIFT is a cheap way to learn how a sophisticated buyer thinks about rubric quality.

How to reach

Research-first; plausible entry via a thoughtful response to RIFT or a proposal on rubrics for visual/design evaluation (speculative).

What we could not establish
  • Location
  • Exact PhD year
  • X handle not verified

Sources

  1. Snorkel AI – Paroma Varma author page
  2. Paroma Varma – Google Scholar
  3. TheSequence interview: Paroma Varma on programmatic data labeling
  4. The Org – Snorkel AI leadership team