Parmar co-authored "Attention Is All You Need," inventing the Transformer.
She co-founded Adept AI Labs and Essential AI after leaving Google.
She now researches reinforcement learning at Anthropic, shaping Claude.
The story of modern artificial intelligence carries an unmistakable Indian imprint, and few names illustrate this reality as convincingly as Niki Parmar. A software engineer who began her career quietly within Google's research division, Parmar went on to co-author one of the most cited papers in computing history. Her work did not merely contribute to progress in machine learning; it redefined the architecture on which nearly every modern language model now runs.
Today, as a member of Anthropic's technical staff, Parmar continues shaping the systems that power tools such as Claude. Her journey, from a college in Pune to the research floors of Google Brain and the founding rooms of two ambitious AI startups, reflects the quiet but formidable rise of Indian talent at the center of the global AI revolution.
Parmar completed her Bachelor of Engineering in Information Technology at the Pune Institute of Computer Technology. An early ambition to study at an IIT did not materialize, a setback that pushed her toward independent learning rather than discouragement. She immersed herself in online courses from AI pioneers Andrew Ng and Peter Norvig, building foundations that would later prove indispensable.
She then pursued a Master of Science in Computer Science at the University of Southern California, graduating in 2015. At USC, she worked with Professor Morteza Dehghani, applying machine learning and large datasets to questions in computational social science. This exposure to interdisciplinary problem-solving shaped her approach to engineering challenges that lay ahead.
Parmar joined Google Research in 2015 and moved into Google Brain in 2017 as a research software engineer. There, she became part of an eight-person team working on sequence transduction models for machine translation. That collaboration produced "Attention Is All You Need," published in 2017, a paper that introduced the Transformer architecture built entirely on self-attention mechanisms.
The paper's numbers spoke for themselves. The proposed model achieved a BLEU score of 28.4 on the WMT 2014 English-to-German translation task, surpassing existing results by more than two points. On the English-to-French benchmark, it established a new state-of-the-art score of 41.0, after training for just 3.5 days on eight GPUs, a fraction of the computational cost used by prior leading models.
This efficiency, paired with the elimination of recurrence and convolution, made the Transformer both faster to train and easier to scale. It became the structural backbone for GPT, BERT, and eventually Claude, cementing Parmar's place among the architects of the generative AI era.
Parmar did not stop at text. She extended self-attention mechanisms into computer vision, developing the Image Transformer for attention-based image generation and contributing to Bottleneck Transformers for vision backbones. She also worked on the Conformer, a model integrating convolutions with transformers for speech recognition. These projects demonstrated that the architecture she helped build could generalize across modalities, from language to sound to visual data.
Parmar left Google in late 2021 to co-found Adept AI Labs alongside Ashish Vaswani and David Luan. As chief technology officer, she helped pursue the concept of "action models": AI systems capable of operating software and browsers on behalf of users, automating repetitive enterprise workflows rather than simply generating text.
In 2023, Parmar co-founded Essential AI with Vaswani, her long-time collaborator from the original Transformer paper. The startup focused on building full-stack AI products designed to automate monotonous, data-heavy workflows for businesses, aiming to compress the time organizations spend on manual operational tasks.
Points that define this entrepreneurial chapter include:
Co-founding Adept AI Labs in 2022 as chief technology officer, focused on enterprise automation
Co-founding Essential AI in January 2023 with Ashish Vaswani
Building products intended to reduce time spent on repetitive, data-driven business processes
Maintaining close ties with fellow Transformer co-authors throughout her startup career
Parmar joined Anthropic as a member of technical staff in December 2024, publicly announcing the move in February 2025. At Anthropic, she works on reinforcement learning research directed at hard exploration tasks, frontier model capabilities, and test-time scaling, alongside broader work in natural language understanding, human feedback, and interpretability.
She has stated that her work contributed to the development of Claude 3.7 Sonnet, the hybrid reasoning model Anthropic released in February 2025, which the company positioned among its most significant releases for coding and complex reasoning tasks. Her arrival coincided with a broader pattern of senior technical leaders, including former founders and chief technology officers from companies such as Adept, Instagram, and Workday, joining Anthropic in similar technical roles rather than executive positions.
Her Google Scholar citation count exceeds 315,000, a figure that reflects the extraordinary influence of her research output across the wider machine learning community, including work built upon by researchers at OpenAI, Google DeepMind, and academic institutions worldwide.
Niki Parmar's career traces a path that few engineers in any country can claim: a single research paper reshaping an entire technological era. From a self-taught student in Pune who missed an IIT admission to a co-author of the Transformer paper, and later a founder of two ambitious AI companies, her trajectory embodies persistence matched with technical rigor.
Her current work at Anthropic places her once again at the frontier of AI development, this time refining the reasoning capabilities of systems built upon the very architecture she helped invent. As India's contribution to global AI leadership continues to draw attention, Parmar's story stands as a compelling reminder that foundational breakthroughs often begin far from Silicon Valley, carried forward by curiosity, discipline, and an unwillingness to accept early setbacks as final outcomes.