Sanket Pathak

About Sanket Pathak

I am a cross-lingual applied and computational linguist with 15+ years of experience in AI-driven language research and large-scale speech and language system development. Currently working as a \'Principal Researcher, Linguistic Solutions & Innovation\', I focus on advancing multilingual ASR and NLP systems for Indian languages, spanning both widely used and under-resourced, typologically diverse language varieties. My work lies at the intersection of linguistic theory, machine learning, and scalable system design, addressing challenges in real-world, high-variability, and code-mixed language environments.My core research focus is on linguistically informed modeling for Automatic Speech Recognition (ASR), particularly through Grapheme-to-Phoneme (G2P) frameworks. I have led phonological analysis, phoneme inventory design, pronunciation modeling, and systematic error diagnostics to develop and optimize ASR pipelines across a wide range of Indian languages. My approach emphasizes cross-lingual transfer, language-aware modeling, and data-centric optimization to improve generalization and robustness in diverse and noisy conditions.Beyond ASR, my work spans Text-to-Speech (TTS), Machine Translation, Conversational AI, and core NLP tasks including transliteration, normalization, morphological analysis, syntactic and semantic annotation, Named Entity Recognition (NER), Word Sense Disambiguation (WSD), and POS tagging. I bring extensive experience in multilingual data strategy, corpus development, annotation framework design, and evaluation methodologies aligned with industrial and research standards. With strong sociolinguistic grounding in Hindi and its regional varieties, including Magahi, Maithili, and Bhojpuri, I am particularly interested in building inclusive, linguistically grounded AI systems and contributing to research that advances the full diversity of Indian languages in real-world applications.

Experience

1. Principal Researcher, Linguistic Solutions & Innovations

   Mihup.ai

   Jun 2025 — Present · Kolkata, IN

   Lead linguistic research and solution design for multilingual ASR and NLP systems, with primary focus on Indian languages across automotive and conversational AI platforms.Drive development of language solutions covering a wide range of Indo-Aryan and Dravidian languages, addressing dialectal variation and real-world speech diversity.Provide expert direction on linguistic modeling, including phonetics, morphology, normalization, and handling of code-mixed and low-resource language scenarios.Collaborate with ML scientists and engineers to align linguistic frameworks with model requirements and system performance objectives.Design and standardize approaches for data annotation, evaluation, and linguistic quality to support scalable ASR development across Indian languages.Conduct detailed error analysis to identify linguistic gaps and support continuous improvement of ASR system performance.Contribute to the development and optimization of pronunciation lexicons and G2P systems for Indian languages.Mentor linguists and language engineers, guiding technical practices and supporting research-driven problem solving.Work with cross-functional teams to ensure alignment between linguistic research, model development, and product requirements.Enabled scalable ASR development across multiple Indian languages by strengthening linguistic data quality, standardization, and evaluation practices.Contributed to measurable improvements in ASR performance through targeted linguistic analysis and data-driven refinements.

Education

• Banaras Hindu University

  Master’s Degree, Applied Linguistics

  2004 — 2006

• University of Lucknow

  Doctor of Philosophy (Ph.D.), Applied and Computational Linguistics

  2006 — 2010