Scaling Intelligence.
Ensuring Reliability.
I build robust, enterprise-scale AI systems and evaluation frameworks. My focus lies at the intersection of production machine learning infrastructure and rigorous linguistic computation.
Background
I grew up in Kannauj and Varanasi, surrounded by the study of Sanskrit. Early exposure to texts like Pāṇini’s Aṣṭādhyāyī—an ancient formal generative grammar—sparked a lifelong fascination with the architecture of language, computation, and structural logic.
I pursued this interest through my BA and MA in Computational Linguistics at BHU, where I was fortunate to receive the University Gold Medal, before completing my PhD at IIT (BHU). My academic research focused on bootstrapping sentiment analysis and detecting bias in low-resource languages, driven by a desire to make computational tools more equitable across diverse dialects.
Transitioning to the industry, I spent five years at Amazon’s Alexa AI, developing Natural Language Understanding (NLU) systems to parse millions of daily conversational interactions across global markets. Later, at Sanas AI, I led the Data and ML Analytics organization, scaling the team to over 40 members and building the automated evaluation pipelines that supported the company’s significant ARR growth.
In 2024, I founded Vaikhari AI to address a critical gap in the enterprise space: providing robust evaluation and governance frameworks for large-scale Generative AI deployments.
Today, I split my time between Bangalore, Varanasi, and Kannauj. Outside of work, I am a student of Hindustani classical music and play a variety of wind and string instruments. For me, exploring the mathematical patterns in rāgas is a natural extension of sequence modelling.
Professional Journey
Founder
Vaikhari AI • Bangalore
2024 – Present
Leading end-to-end product, research, and customer discovery for an enterprise AI evaluation platform. Designed the CORE benchmark frameworks for ontological reasoning, robustness, and real-world task performance at scale. Built product strategy and GTM positioning for AI governance.
Director of ML Analytics
Sanas AI • Bangalore
2023 – 2024
Founded and scaled the Data and ML Analytics org to 40+ HC, supporting significant ARR growth. Led quarterly strategic planning, managing multi-million-dollar budgets. Built the Model Factory for automated data flow and evaluation.
Lead NLU Engineer
Amazon (Alexa AI) • Bangalore
2019 – 2023
Led NLU engineering for Alexa AI, managing global cross-functional teams of engineers and scientists. Shipped ML systems supporting millions of daily user interactions across global markets, improving customer experience through automation and model optimization.
Senior Researcher & TA
IIT (BHU) • Varanasi
2017 – 2019
Taught DSP/NLP courses and led end-to-end data, modeling, and evaluation for NLU research for applied projects. Mentored B.Tech students while pursuing computational linguistics research.
Selected Research
Alexa teacher model: Pretraining and distilling multi-billion-parameter encoders for NLU systems
FitzGerald J, Dwivedi S, et al.
Navigating Linguistic Diversity: ICL and Prompt Engineering for Subjectivity Analysis in Low-Resource Languages
Dwivedi, S., Ghosh, S. & Dwivedi, S.
Breaking the bias: Gender fairness in LLMs using prompt engineering and in-context learning
Dwivedi, S., Ghosh, S. & Dwivedi, S.
Binary classifier for identification of stammering instances in Hindi speech data
Dwivedi, S., Ghosh, S. & Dwivedi, S.
Research & Focus Areas
My work inherently spans across classical linguistics and modern machine learning infrastructure. Here are the intersections I spend my time exploring, building, and deploying.
AI Evaluation & Benchmarking
AI Reliability & Governance
Conversational AI
Generative AI
Bias in LLMs
Low-Resource Language NLP
Speech AI
Sanskrit Computational Linguistics
Paninian Grammar
Equitable AI
Engineering Stack
LLMs, Evaluation Frameworks, A/B Testing, State Space Models, Transformers
AWS (Bedrock, Lex), GCP (Vertex AI), Large-scale ML Pipelines
Python, SQL, Bash
In The Field
Replace Image in WP
AI Summit 2024
Keynote Speaker
Replace Image in WP
Low-Resource NLP
Panel Discussion
Replace Image in WP
Hindustani Music
Live Performance
Let’s Connect
Whether you are interested in discussing AI Governance, looking for collaboration on Low-Resource NLP, or just want to chat about the grammar of Panini, my inbox is open.