Ms. Shifali Agrahari is an accomplished researcher in Natural Language Processing (NLP), Machine Learning (ML), and Artificial Intelligence (AI), currently engaged in advanced research projects focused on AI-generated text detection, multilingual NLP, and graph-based deep learning models. She holds an M.Tech degree in Computer Science with a specialization in Artificial Intelligence, having transitioned from a B.Tech in Chemical Engineering, and has demonstrated exceptional performance in computer science research. She also completed a six-month research internship at Samsung Research Institute India–Bangalore, further strengthening her expertise in applied AI.

Her M.Tech thesis centered on enhancing sentiment classification for Hindi data through novel text-entailment-based techniques using LSTM and consistency regularization. Her research and project experience reflect deep technical expertise in transformer architectures (BERT, RoBERTa, T5, Pegasus, GPT), graph neural networks (GCN, GNN), and large-scale dataset creation for multilingual and multimodal tasks. She has also participated in prestigious initiatives such as the Amazon ML Summer School, where she contributed to human gaze-driven egocentric video understanding using attention-based GCN models.

Research Contributions

Ms. Agrahari’s scholarly contributions have been recognized in top-tier international conferences and journals including COLING-GenAI 2025, ACM 2024, and IEEE Transactions (under review).
Her research primarily addresses pressing challenges in AI-generated text detection, multilingual LLM evaluation, authorship attribution, and fake news detection.

Notable works include:

• Can You Really Trust That Review? A Prototypical Few-Shot Method and Multi-Domain Benchmark for Detecting AI-Generated Reviews (IJCNLP-AACL 2025, Accepted) – Introduced a prototypical network-based approach for AI-generated text detection across multiple domains.

• Multilingual MGT Detection (COLING-GenAI 2025) – Developed MLDet, achieving strong F1 performance across languages using cross-lingual adaptation.

• Guardians of Academic Integrity (COLING-GenAI 2025) – Proposed a fusion framework integrating linguistic and stylometric cues to detect AI-generated academic writing.

• Text Authorship Attribution (ACM 2024) – Pioneered comparative stylometric analysis across LLMs like GPT, Gemini, and LLaMA to distinguish human and machine authorship.

• Fake News Detection Using Hashtag Context (Pattern Recognition Letters, Accepted) – Proposed HCFND, a hashtag-aware model improving context-driven fake news detection.

Projects & Technical Achievements

• Distinguishing AI-Generated vs Human Reviews: Achieved 85.6% accuracy using RoBERTa and BERT classifiers across multiple domains (hotel, product, book, movie).

• Tweet Geolocation Prediction using GNN: Designed a GCN + FNN hybrid model improving accuracy from 69% to 81%.

• Hindi Sentiment Analysis (M.Tech Project): Achieved 76% accuracy using a joint objective with consistency regularization.

• Human-Gaze-Enhanced Video Recognition (Amazon ML School): Advanced egocentric activity recognition by fusing gaze data with optical flow-based GCN.

Scientific Impact & Innovation

Her body of work demonstrates:

• Strong multilingual and cross-domain generalization of AI models.

• Novel dataset creation benefiting the wider research community.

• Integration of linguistic theory, deep learning, and knowledge-graph reasoning for real-world problems.

• Leadership in developing reproducible and ethical AI systems for academic integrity and information authenticity.

Eligibility for Women Researcher Award

Given her significant contributions to AI and NLP research, multiple international publications, and innovation in multilingual and ethical AI, Ms. Shifali Agrahari stands out as an exemplary candidate for the Women Researcher Award (Scientific Laurels). Her work not only advances the state of research in machine-generated text detection and multilingual understanding but also inspires the next generation of women researchers in AI.