Currently at Apple · Building Apple Intelligence

AmanSaini.

ML Engineer at Apple.

Generative AI · LLM Post-training · Reinforcement Learning · LLM Alignment · Natural Language Generation · Classic ML

Get in touch→LinkedIn ↗Google Scholar ↗
Aman Saini
01 / Work

Experience.

Apple

Apple

Apr 2025 — Present
Machine Learning Engineer · Vancouver, BC
  • Building World-Class Question Answering for Siri.
  • Training and optimizing GenAI models using advanced post-training methods (DPO, Online RL).
  • Designing verifiable reward pipelines with rubrics and improving LLM alignment using better Reward Models.
  • Developing scalable evaluation and training frameworks for high-quality generative AI experiences.
Grammarly

Grammarly

Oct 2023 — Apr 2025
Staff ML Engineer · Vancouver, BC
  • Fine-tuned LLMs (Llama, GPT) for fine-grained writing assistance; instruction-tuned Llama 3.1 70B/8B teacher models.
  • Built evaluation pipelines using LLM-as-a-Judge; scaled synthetic data generation for internal LLMs.
  • Published multi-task Ukrainian text-editing work; released models and datasets on Hugging Face.
  • Created a Python library for automatic prompt optimization with target reward metrics.
Twitter / X

Twitter

May 2021 — Jan 2023
Senior ML / NLP Engineer · Seattle, WA
  • NLP Signals (Cortex): End-to-End NLP models and signals used across Home Timeline, Notifications, and Trends.
  • Entity Linking: Built Encoder-only Transformer models for detection, candidate generation and ranking to identify and link Named Entities with Tweets. Dataset released at NeurIPS 2022.
  • NER: Replaced Bi-LSTM with multilingual BERT using subword masking, token features, weak labels, and distillation, resulting in a 13.5% lift in offline F1 score.
  • Tweet Representation: Multi-task model for language, topical, and engagement prediction; published in DL4SR 2022.
  • Improved notification seed-Tweet quality; led NER adoption and ML code guidelines.
Microsoft

Microsoft

Jan 2016 — Apr 2021
Senior ML Engineer · Seattle, WA
  • Bing Ads (AI + Research): Improved ad experience and launched new ad products.
  • Dynamic Search Ads: Landing-page extraction and Transformer-based generation — +1.5% US daily RPM, 2%+ international RPM.
  • Web Page Similarity Graph: TwinBERT / USE embeddings with HNSW / NSG ANN for next-link prediction.
  • Dynamic links with ads: DeepXML categories + Ads RoBERTa scoring — +1.5% US CTR.
  • Landing-page summarization (BERTSUM) used in ad title assets; latency tooling adopted across services.
Microsoft

Microsoft India

Jul 2013 — Dec 2015
Software Engineer · Bangalore, India
  • Rich Ads experience (RnR): Online infra and rich ad experiences for non-US markets.
  • Ads-Composition: Unified, parallelized pre-serve pipeline for ads and decorations — lower latency, faster experimentation.
  • Related Product Annotation: Surfaced relevant products for retail queries — higher CTR on text ads and product links.
  • Latency tooling: Utilities to analyze critical latency paths, reused across services.
02 / Research

Published work.

arXiv 2024

Spivavtor: An Instruction-Tuned Ukrainian Text Editing Model ↗

Aman Saini, Artem Chernodub, Vipul Raheja, Vivek Kulkarni

Spivavtor is a Ukrainian-focused instruction-tuned text-editing model and dataset, adapting the CoEdIT framework to Ukrainian. We describe dataset construction, model training, and evaluation.

NeurIPS — Datasets & Benchmarks, 2022

TweetNERD — End-to-End Entity Linking Benchmark for Tweets ↗

Shubhanshu Mishra, Aman Saini, Raheleh Makki, Sneha Mehta, Aria Haghighi, Ali Mollahosseini

A large-scale benchmark for entity extraction and linking on Twitter spanning 2010–2021 with three tasks: NER, Entity Linking with gold spans, and End-to-End Entity Linking.

arXiv 2025

APIO: Automatic Prompt Induction and Optimization for Grammatical Error Correction and Text Simplification ↗

Artem Chernodub, Aman Saini, Yejin Huh, Vivek Kulkarni, Vipul Raheja

APIO is a two-stage approach that induces task-specific prompts from examples and optimizes them for text-editing tasks, attaining strong SARI scores on ASSET-Test.

03 / Earlier

Internships.

Jan — Jun 2013

Amazon

Software Engineer Intern · Bangalore, India
  • · Worked with the Amazon Pricing team to predict shipping prices for products on amazon.com.
  • · Implemented the End-to-End Pricing Attribute Prediction Service using Amazon Simple Workflow Service to automate the ML model-building process.
May — Jul 2012

University of Victoria

Research Intern · Victoria, Canada
  • · Replaced Java with the bidirectional language Boomerang in a high-confidence medical data device (MirthConnect) to guarantee well-behaved channels.
May — Jul 2011

Indian Institute of Remote Sensing

Research Intern · Dehradun, India
  • · Worked with the Weather Research and Forecasting (WRF) modeling team to deploy WRF models on Linux, enabling real-time weather prediction.
04 / Education

Education.

Birla Institute of Technology and Science (BITS) Pilani ↗

2009 — 2013
B.E. in Computer Science · CGPA 9.54 / 10

Let's talk.

contact@aman-saini.com→
Vancouver, BC, Canada