CV

Ph.D.

George Mason University

Computer Science

• Trustworthy and Interpretable AI
• LLM Explainability
• LLM Agents

M.S.

University of Virginia

Computer Science

• Machine Learning
• Neural Contextual Bandits

B.S.

Xidian University

Computer Science

CAAC: Confidence-Aware Attention Calibration to Reduce Hallucinations in Large Vision-Language Models

WACV 2026

Presents an attention calibration method that mitigates hallucinations in large vision-language models by accounting for model confidence.

Making Sense of LLM Decisions: A Prototype-based Framework for Explainable Classification

AAAI 2026 (Oral)

Develops a prototype-based framework to explain classification decisions in large language models with transparent, interpretable reasoning.

CORTEX: Collaborative LLM Agents for High-Stakes Alert Triage

NeurIPS 2025 LAW Workshop

Proposes a collaborative LLM-agent framework for security operations center alert triage with interpretable decision-making.

ProtoLens: Advancing Prototype Learning for Fine-Grained Interpretability in Text Classification

ACL 2025 (Main)

Introduces ProtoLens, improving interpretability in text classification through prototype learning and fine-grained analysis of linguistic features.

VIGNETTE: Socially Grounded Bias Evaluation for Vision-Language Models

In Submission

Proposes a framework to evaluate and mitigate social bias in vision-language models through grounded scenarios.

Neural Symbolic Logical Rule Learner for Interpretable Learning

ICLR 2026 (Under Review)

Combines neural representation learning with symbolic logical reasoning for transparent decision boundaries.

An Empirical Study of Neural Contextual Bandit Algorithms

M.Sc. Thesis, University of Virginia

A comparative analysis of contextual bandit algorithms with neural approximators in recommendation and decision-making tasks.

Evidence-Attribution Reinforcement Learning (EA-RL)

Develops reinforcement learning methods rewarding LLMs for using correct evidence, not only producing correct answers.

• Designs multi-agent distillation with explicit evidence attribution.
• Builds on CoA for faithful, verifiable single-agent reasoning.

NeuroSymbolic Autoencoder for Interpretable Recommendation

Integrates neural and symbolic reasoning for explainable recommendation systems using rule-based latent spaces.

• Employs Rule Network as encoder-decoder for transparent RecSys.
• Targets SIGIR 2026 for publication.

GenAI Engineer Intern

GoEngage

Implemented semantic retrieval and built an agentic chatbot that interfaces with APIs to generate analytical reports.

• Implemented semantic search outperforming keyword baselines.
• Built a reasoning-capable chatbot integrated with backend APIs.

AI Agents Developer Intern

Fluency Security

Developed multi-agent LLM systems for SOC alert triage; designed the CORTEX architecture for explainable collaboration among LLM agents.

• Proposed and built CORTEX multi-agent LLM framework for alert triage.
• Curated analyst-trace dataset (10+ scenarios) with tool outputs.
• Improved actionable F1 to 0.78 (+0.12) and reduced false-positive rate to 14.2%.
• Paper accepted to NeurIPS 2025 LAW Workshop.