Shivangi Tripathi PhD Student · Texas State University

3rd-Year PhD · Computer Science

Hi, I'm Shivangi.

I'm a doctoral researcher at Texas State University, working in the Human AI Synergy Lab under Dr. Heena Rathore. My research centers on making large language models and vision-language models more trustworthy — from detecting and mitigating hallucinations to understanding failure modes under adversarial attacks and building reliable evaluation frameworks for real-world deployment.

Trustworthy AI LLM / VLM Reliability Hallucination Detection & Correction Adversarial Robustness Evidence-Grounded Generation
Actively seeking Summer 2026 research internships in LLM/VLM reliability, hallucination detection & mitigation, robust evaluation, and agentic AI systems.
Shivangi Tripathi
Lab Human AI Synergy Lab
Location San Marcos, TX, USA

Recent Publications

Selected peer-reviewed publications and conference proceedings.

  1. Detecting and Correcting Hallucinations in Paragraph-Level Text with Ensemble-Based Evaluation IEEE DISTILL, 2025

    This work proposes an ensemble-based evaluation and correction pipeline that combines BERT-style classifiers, semantic similarity, NLI, and LLM-as-judge signals to detect hallucinations in paragraph-level generations and guide targeted revisions, improving factual consistency of long-form outputs. [web:1][web:20]

  2. Paragraph-Level Hallucination Detection for LLMs in Networked Systems Proceedings of IEEE CCNC, 2026

    This project models LLM deployments in networked and edge environments and introduces paragraph-level hallucination detection methods and metrics tailored to bandwidth, latency, and resource constraints, enabling more trustworthy text generation in realistic system pipelines. [web:1]

  3. REVISE: A Framework for Paragraph-Level Misinformation Correction in Large Language Models Proceedings of CIOCSE, 2025

    REVISE is a structured framework that decomposes paragraphs into verifiable claims, generates verification questions, retrieves external evidence, and uses a gated decision mechanism to drive LLM-based rewriting, enabling systematic correction of misinformation while preserving coherent, context-aware narratives. [web:1][web:22]

  4. Assessing Hallucination in LLMs under Adversarial Attacks Proceedings of MOBISECSERV, 2024

    This study evaluates open-source models such as Vicuna-7B and LLaMA2-7B-chat under prompt-based adversarial attacks, quantifying attack success rates and loss shifts to show how weak semantic and out-of-distribution perturbations significantly increase hallucination incidence. [web:16][web:2]

Education

Academic training across AI, signal processing, and engineering.

Ph.D. in Computer Science

Texas State University · Aug 2023 – Present (Expected 2027)

  • Lab: Human AI Synergy Lab
  • Advisor: Dr. Heena Rathore
  • Focus: Machine Learning, LLM Security, Hallucination Detection, Trustworthy NLP

M.S./M.Tech in Electrical Engineering

IIT Jodhpur · Jul 2021 – Jul 2023

  • Focus: Cyber-Physical Systems, 6G Networks, ML for Communications

B.Tech in Electronics & Communication Engineering

GGSIPU, Delhi · Aug 2017 – May 2021

Research Experience

Key roles and contributions.

Doctoral Instructional Assistant / Researcher

Human AI Synergy Lab, Texas State University · San Marcos, TX

Aug 2023 – Present
  • Designed and evaluated hallucination detection and mitigation pipelines with measurable reliability improvements.
  • Built reproducible ML experiments in Python using PyTorch, TensorFlow, and Hugging Face for controlled benchmarking.
  • Analyzed adversarial robustness and failure modes to inform evaluation criteria and mitigation strategies.

Research Assistant

Wireless Communications Lab, IIT Jodhpur · Jodhpur, India

Jul 2021 – Jul 2023
  • Developed and simulated joint resource allocation algorithms for 6G networks under mobility and latency constraints.
  • Implemented MIMO configurations and quantified Doppler impacts using signal processing and linear algebra.
  • Designed optimization strategies for mission-critical Mobile Edge Computing (MEC) applications.

Selected Projects

High-impact research and engineering highlights.

Paragraph-Level Misinformation Correction for LLMs

Jun 2024 – Dec 2025

  • Built a pipeline: sentence decomposition → verification QG → evidence retrieval → context-aware revision.
  • Achieved 76.8% correction accuracy while maintaining FactCC > 0.97 across 1,400+ passages.

Hallucination Under Adversarial Attacks

Aug 2023 – May 2024

  • Implemented prompt-based attacks to induce hallucinations under controlled evaluation settings.
  • Observed attack success rates of up to 92.53% (weak semantic) and 88.68% (OoD).

Extractive Question Answering Models

Sep 2024 – Dec 2024

  • Implemented LSTM, custom Transformer, and fine-tuned BERT in PyTorch on SQuAD and Natural Questions.
  • Achieved up to 83% F1, outperforming TF-IDF baselines by over 20%.

Resource Allocation for 6G Networks

Jun 2022 – Jul 2023

  • Modeled dynamic uplink/downlink resource sharing in MATLAB/Simulink for high-mobility scenarios.
  • Improved BER by 30% and spectral efficiency by 25%.

Hobbies

Outside of research.

When I'm not in the lab, I enjoy staying active and exploring new experiences. I love hiking, playing badminton, and attending dance workshops — activities that keep me energized and present.

Lately, I've been developing a new interest in landscape photography — capturing natural light, open skies, and quiet outdoor spaces.

Contact

For research collaboration, internship opportunities, or academic discussions.