Oct 2025 – Nov 2025 · 1 mo · Ireland

Oct 2024 – Jul 2025 · 9 mos · Dublin, County Dublin, Ireland

• I worked on evaluating and fine-tuning large language models using Reinforcement Learning from Human Feedback (RLHF), focusing on improving alignment, reasoning, and factual accuracy. I designed and deployed adversarial prompts to intentionally provoke model failures, particularly in physics, mathematics, and coding tasks, helping uncover edge cases and reasoning flaws. My work involved analysing model outputs, ranking completions, and providing detailed human feedback to guide iterative training cycles. By probing the models across complex STEM domains, I contributed to enhancing their robustness, safety, and overall performance, while also informing data curation strategies and model update priorities.

Oct 2023 – Dec 2025 · 2 yrs 2 mos

Oct 2023 – Aug 2024 · 10 mos · Coimbatore, Tamil Nadu, India · On-site

Nov 2022 – Dec 2023 · 1 yr 1 mo

Dec 2020 – Nov 2022 · 1 yr 11 mos

Jan 2020 – Dec 2020 · 11 mos

Mar 2021 – Jan 2024 · 2 yrs 10 mos · Rochester, New York, United States · Remote

• According to the article sourced by The World Bank that the internet traffic was estimated to be more than 3 zettabytes i.e (3 trillion) which is an unbelievable growth in network traffic resulting in data loss and delayed transmission of information. To tackle this issue that has been predominantly rising, A protocol that is being developed to address the current issues faced in networking traffics is called MNLR protocol (Multi Node Label Routing) also called EIBP (Expedited Internet Bypass Protocol). So whenever a particular node or a link fails this protocol immediately takes a different route. By operating beneath the current internet protocols, the new protocol ensures the uninterrupted operation of regular internet traffic. Thus offering a way more efficient transfer of infos and data in the network.