Research

Academic Publications

TEEMIL: Towards Educational MCQ Difficulty Estimation in Indic Languages

The 31st International Conference on Computational Linguistics (COLING 2025)

Ravikiran, M., Vohra, S., Verma, R., Saluja, R., & Bhavsar, A. (2025). TEEMIL: Towards Educational MCQ Difficulty Estimation in Indic Languages. The 31st International Conference on Computational Linguistics (COLING 2025). ACL Anthology.

You Reap What You Sow—Revisiting Intra-class Variations and Seed Selection in Temporal Ensembling for Image Classification

International Conference on Frontiers in Computing and Systems (COMSYS 2023)

Ravikiran, M., Vohra, S., Nonaka, Y., Kumar, S., Sen, S., Mariyasagayam, N., & Banerjee, K. (2023). You Reap What You Sow—Revisiting Intra-class Variations and Seed Selection in Temporal Ensembling for Image Classification. In Proceedings of International Conference on Frontiers in Computing and Systems (pp. 73-82). Springer, Singapore.

(Manikandan Ravikiran and Siddharth Vohra—Both authors contributed equally. Names are ordered alphabetically)

Investigating the Effect of Intraclass Variability in Temporal Ensembling

arXiv preprint (2020)

Vohra, S., & Ravikiran, M. (2020, August 21). Investigating the effect of intraclass variability in temporal ensembling. arXiv.org

Technical Blogs

Using WAF with App Runner in Copilot

February 23, 2023

Vohra, S. (2023, February 23). Using WAF with app runner in copilot. AWS Copilot CLI Blog

Research Experience

Independent Research

June 2020 - Present

• Conducted independent research with mentors from Hitachi R&D and Thoughtworks, spanning traditional machine learning, large language models (LLMs), and computer vision
• Co‑authored a COLING 2025 paper introducing a multiple‑choice question benchmark for Hindi and Kannada educational texts and providing LLM‑based baseline difficulty‑estimation models.
• Co‑authored a AIED 2025 submission investigating whether semantic similarity between answer options can predict question difficulty
• Benchmarked Temporal Ensembling semi‑supervised models across diverse vision datasets, analysing intra‑class variability impacts in collaboration with Hitachi R&D; findings published at COMSYS 2023
• Initiated and led a project to benchmark coding‑question difficulty using LLMs, developing a BERT‑style model for automatic programming‑problem complexity estimation.
• Explored computer‑vision techniques for pedestrian‑attribute recognition and person re‑identification, evaluating multiple network architectures to improve identification accuracy in security applications.
• Built reproducible pipelines to fine‑tune GPT, Claude, Llama and Gemma models with LoRA/QLoRA, enabling single‑GPU experimentation.