Mahek Vora

Research

Publications

I've been lucky to work with research groups at CMU, UPenn, UNSW and IIT Guwahati on problems spanning structural biology, NLP and LLMs.

DUIL model architecture
NeurIPS 2026 · Poster

DUIL: Deep Unsupervised Inverse Learning for in situ Macromolecular Morphology Identification

Carnegie Mellon University · Advisor: Prof. Min Xu

  • Designed and trained an unsupervised autoencoder with a custom Multi-Choice Learning module that disentangles pose and semantic embeddings of protein structures.
  • Outperforms RELION by 40% ARI on low-SNR Cryo-ET data.
  • Identified a previously unresolved ATP synthase structure in the thylakoid membrane of yeast using this method.
Paper
TempTabQA dataset domains and results
EMNLP 2023 · Main

TempTabQA: Temporal Question Answering for Semi-Structured Tables

University of Pennsylvania · Advisors: Prof. Vivek Gupta & Prof. Vivek Srikumar

  • Co-authored an 11,545-pair QA dataset spanning 1,208 Wikipedia infobox tables across 90+ domains.
  • Exposed a 13.5+ F1-point temporal-reasoning gap between top LLMs and human performance.
Paper
Sentiment analysis pipeline
BDCC 2026 · Preprint

Large Language Models for Bible Sentiment Analysis: Sermon on the Mount

University of New South Wales · Advisor: Prof. Rohitash Chandra

  • Applied S-BERT embeddings and the AFINN lexicon to quantify sentiment variance across 5 Bible translations.
  • Analysed verse-by-verse polarity scores to draw historical inferences.
Paper
AmalREC 5-stage pipeline
Preprint

AmalREC: A Dataset for Relation Extraction and Classification Leveraging Amalgamation of LLMs

IIT Guwahati · Advisor: Prof. Ashish Anand

  • Designed a 5-stage pipeline for relation extraction and classification data generation.
  • Released a dataset spanning 255 relation types with ~200k sentences.
Paper