Jul 2023 – Nov 2024 · 1 yr 4 mos · Gurugram, Haryana, India · On-site

• Identified critical theoretical gaps in Tokenization strategies for Large Language Models (LLMs), validated findings by Pre-Training multiple models across 4+ language groups, leading to a paper submission at IEEE COLING 2025.
• Established the AI front for the Crisis Monitoring Product within Sprinklr’s Social Insights suite, leveraging social-media data across multiple channels to enable real-time identification and tagging of crisis events from more than 2,000 categories spanning over 10 industries, all powered by LLMs.
• Led the development of AI pipelines for Sprinklr’s Contact Driver Intent module, part of the Conversational Analytics suite, by packaging open-source fine-tuned models and deploying them using VLLM, resulting in a horizontally scalable, fault-tolerant, high-throughput solution shipped to Netflix as the first client.

Jan 2023 – Jul 2023 · 6 mos · Gurugram, Haryana, India · On-site

• Experimented with various Large Language Models such as Flan-T5, Dolly, OPT, and Vicuna etc; to develop a training Data Generator. This initiative aimed to reduce annotation costs by generating synthetic training data, leveraging the capabilities of these models.
• Collaborated with clients such as Etihad and WestJet to expand the Airline Vertical's Intent Library. Employed various community detection algorithms to refine class labels and achieve desired levels of granularity, ensuring a more effective intent classification system.

May 2022 – Jul 2022 · 2 mos

• Engineered a custom encoder head and pre-processed diverse datasets, improving Intent Classification accuracy across 19 industries and 200+ categories.
• Enhanced the RoBERTa model via integrated novel pre-training techniques, custom tokenization, and architectural modifications, contributing to extensive evaluation of transformer-based models.
• Realized a substantial increase in accuracy to 82.59% on the intent classification model through rigorous real-world testing with high-profile clients like Adobe.

Feb 2022 – Apr 2022 · 2 mos

• I did this to earn some extra bucks in college. I used to solve competitive programming problems on camera and was paid some money for this

Aug 2021 – May 2022 · 9 mos · Prayagraj, Uttar Pradesh, India

• Supervisor -- Dr Rahul Kala
• Applied interdisciplinary research skills in robotics, automation, and logistics to advance several projects, resulting in a publication at IEEE CASE in 2022.
• Developed and open-sourced a modular warehouse simulator, employing innovative techniques in warehouse transportation and congestion management.
• Teamed up with globally renowned institutions like the University of Illinois and IIT Delhi, contributing to a research paper submitted to IEEE ROMAN 2023 while facilitating advancements in warehouse automation and robot motion planning.