Ashish Kumar.

2023 – 2025 / Ola Krutrim

Director, Applied AI

At Ola’s Krutrim, I led applied AI work across large language model fine-tuning, alignment and production readiness. This included Krutrim Spectre v1 and v2 and the Krutrim chat assistant.

I also led the development of enterprise AI applications, including an agent builder platform and contact center AI. The role connected language model research with the applications and infrastructure needed to use those models in products.

2021 – 2023 / Salesken

Vice President, Data Science

I led AI strategy for a sales intelligence platform, working across conversation analysis, real-time call guidance, lead scoring and pipeline intelligence.

The research and engineering work included domain-aligned language models, model quantization, speech transcription and inference efficiency. I led a team of 35 across research and applied machine learning, with research moving into product on six-week cycles.

My publicly available models on query well-formedness and paraphrase diversity also reflect this interest in evaluating the quality of language used in conversational systems.

2019 – 2021 / Reliance Jio Haptik

Assistant Vice President

My work at Reliance Jio Haptik focused on conversational AI, natural language understanding and the AI research function. This included extending the platform to 12 Indic languages through transliteration, language identification and multilingual entailment.

I worked on intent recognition, automated evaluation, intent discovery and training-data generation. The platform work also extended from chat to voice and enterprise document workflows, including speech-recognition correction and question answering over unstructured documents.

2016 – 2019 / Active.Ai

Vice President, AI

I founded and led the AI research and development function, setting technical direction and owning the AI product roadmap. I architected Triniti, the company’s conversational AI platform, from its foundations through production.

That work covered language understanding, generation and inference, machine comprehension, dialogue management, multi-turn context and question answering. The team’s machine comprehension work reached ninth place on the SQuAD 1.1 benchmark.

Earlier work / Data science and machine learning

Search, recommendations and customer behaviour

At IGP.com, I led data science work on gift discovery, recommendations, intelligent search, ranking, image tagging and customer segmentation. At attune, I developed machine learning capabilities for retail recommendations, customer behaviour modelling, clienteling and social media analytics.