Priya Sharma

Data Scientist · ML Engineer

Staff-level data scientist with 7+ years building ML systems at scale. Shipped recommendation engines to 400M users at Spotify, built LLM evaluation pipelines at OpenAI, and published 4 peer-reviewed papers.

✉ priya.s@email.com 🔗 linkedin.com/in/priyasharma 📄 scholar.google.com 📍 San Francisco, CA

Experience

Staff Data ScientistFeb 2022 – Present

OpenAI · San Francisco, CA

Built fine-tuning evaluation pipeline reducing benchmark variance by 18%. Designed human preference data collection framework processing 100k+ comparisons/week. Co-authored 2 internal papers on RLHF evaluation adopted for GPT-4 training.

PyTorchRLHFEvaluation
Senior ML EngineerJul 2019 – Jan 2022

Spotify · New York, NY

Owned recommendations algorithm serving 400M users. Led 12% CTR improvement via two-tower neural retrieval model. Reduced training costs 40% through mixed-precision training. Built real-time feature store serving <10ms p99.

TensorFlowSparkAirflowRedis
Data Science InternJun – Aug 2018

Google Brain · Mountain View, CA

Research intern on AutoML team. Proposed and prototyped neural architecture search heuristic incorporated into NAS-Bench-301 benchmark.

JAXAutoML

Selected Publications

Scaling Laws for Reward Model Calibration in RLHF

NeurIPS 2023 · Best Paper Award

Sharma P., et al. · 312 citations · arXiv:2312.04567

Efficient Two-Tower Retrieval at Scale: Lessons from 400M Users

RecSys 2021 · Industry Track

Sharma P., Kumar A., et al. · 189 citations

Beyond Accuracy: Preference-Aware Evaluation for Generative Models

ICML 2024

Sharma P., et al. · 94 citations · arXiv:2401.09812