Krishna Khadka

Researcher & builder · UT Arlington

Making AI easier
to understand.

I study how machine learning systems make decisions and how to improve them when they fail. My work connects explainability, software testing, and model repair.

Explore my work Google Scholar

PhD candidate in Computer Science · Co-founder, Bhasha Tech

Krishna Khadka
Krishna KhadkaDallas–Fort Worth, TexasTrustworthy AI
Research in collaboration with NIST
8peer-reviewed papers
70+research citations
112of 30,709 submissions
NeurIPS 2026 Oral

01 / Selected research

Inside the black box.

Three ways I approach trustworthy AI: understand a prediction, explain a decision, and repair a recurring failure.

NeurIPS 2026 · Oral

DD-CAM

112 of 30,709submissions selected for an oral presentation
Approximately 0.36% of submissions

What does a vision model actually need to see?

I use delta debugging to identify a minimal sufficient set of image regions that supports a model's prediction.

40% improvement in medical image localization.

Read the paper Publication details

KDD 2026 · Oral

ABLE

How can a local explanation stay faithful?

Adversarial pairs bracket a decision boundary, providing a foundation for interpretable local surrogate models.

State-of-the-art fidelity for local explanations.

Explore the code Publication details

AAAI 2027 · Under review

DSR

What if a model keeps making the same mistake?

Decision signatures guide post-hoc repair of repeated classification failures across vision, language, and tabular data.

12 datasets. 74.9% average worst-group accuracy.

More about the results

DSR achieved the best average worst-group accuracy (74.9%) and overall accuracy (82.8%) among seven post-hoc repair methods in the manuscript's evaluation. This work is under review.

Explore the complete publication list 8 peer-reviewed papers + 1 manuscript under review

8 peer-reviewed papers, 1 manuscript under review, 70+ citations

2026

  • DD-CAM: Minimal Sufficient Explanations for Vision Models Using Delta Debugging
    K. Khadka, Y. Lei, R.N. Kacker, D.R. Kuhn
    Advances in Neural Information Processing Systems (NeurIPS) 2026 (Oral; 112 of 30,709 submissions)
    Improved medical image localization by 40%; arXiv:2602.19274

  • ABLE: Using Adversarial Pairs to Construct Local Models for Explaining Model Predictions
    K. Khadka, S. Shree, P. Budhathoki, Y. Lei, R. Kacker, D.R. Kuhn
    32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining 2026 (Oral)
    Achieved state-of-the-art fidelity on local explanations using a novel adversarial bracketing technique

  • A Combinatorial Approach to Synthetic Data Generation for Machine Learning
    K. Khadka, J. Chandrasekaran, Y. Lei, R. Kacker, D.R. Kuhn
    SN Computer Science, vol. 7, no. 59, 2026

  • TabKD: Tabular Knowledge Distillation through Interaction Diversity of Learned Feature Bins
    S.N. Pereira, K. Khadka, Y. Lei
    International Joint Conference on Artificial Intelligence (IJCAI) 2026

Under Review

  • DSR: Post-Hoc Repair of Repeated Classification Failures with Decision Signatures
    K. Khadka, S.N. Pereira, Q. Wei, Y. Lei
    Under review at AAAI 2027
    Achieved the best average worst-group accuracy (74.9%) and overall accuracy (82.8%) among seven post-hoc repair methods across 12 vision, language, and tabular datasets

2024

  • Constructing Surrogate Models in Machine Learning Using Combinatorial Testing and Active Learning
    S. Shree, K. Khadka, Y. Lei, R.N. Kacker, D.R. Kuhn
    39th IEEE/ACM International Conference on Automated Software Engineering (ASE)

  • A Combinatorial Approach to Hyperparameter Optimization – Distinguished Paper Award Candidate
    K. Khadka, J. Chandrasekaran, Y. Lei, R.N. Kacker, D.R. Kuhn
    IEEE/ACM 3rd International Conference on AI Engineering (CAIN)

  • Assessing the Degree of Feature Interactions that Determine a Model Prediction
    K. Khadka, S. Shree, Y. Lei, R.N. Kacker, D.R. Kuhn
    IEEE International Conference on Software Testing, Verification and Validation Workshops (ICSTW)

2023

  • Synthetic Data Generation Using Combinatorial Testing and Variational Autoencoder
    K. Khadka, J. Chandrasekaran, Y. Lei, R.N. Kacker, D.R. Kuhn
    IEEE International Conference on Software Testing, Verification and Validation Workshops (ICSTW)

02 / Building in the real world

AI that helps
people find their voice.

At Bhasha, I build AI for language learning: personalized lessons, pronunciation feedback, and conversations that help learners practice English through their native language.

Co-founder · Personalization, speech & conversational AI

70,000+learners
130+countries

My work connects speech recognition, language models, and adaptive feedback into a learning experience on iOS and Android.

Meet Bhasha App Store ↗Google Play ↗
Bhasha app showing lesson categories with native-language guidance
Bhasha app showing an AI role-play voice conversation
Real screens from Bhasha · Learn, practice, speak

03 / A little about me

Research with
a builder's instinct.

I bring a software engineer's perspective to machine learning. Testing, verification, and debugging give me practical tools to ask better questions about AI and to improve systems when their answers fall short.

At UT Arlington, I research trustworthy AI in collaboration with NIST. Alongside that work, I teach computer science and build products that put AI in people's hands.

Understand. Test. Improve.

Education

PhD in Computer Science

The University of Texas at Arlington
Machine Learning & AI · Expected December 2026 · GPA 4.0

BS in Software Engineering, UTA · 2021
Summa Cum Laude · GPA 4.0

Teaching

Making complex ideas approachable.

Graduate Teaching Assistant at UTA since Fall 2024, and guest lecturer in machine learning at Winston-Salem State University in November 2025.

Recognition

Research, teaching & entrepreneurship.

UTA's 2026 Outstanding Graduate Teaching Assistant Award, Hesed Endowed Scholarship, and MavPitch recognition for Bhasha.

Giving back to the field.

Peer review & academic service

  • KDD 2025: Reviewer
  • IJCAI 2026: Reviewer
  • ANTIC 2025: Reviewer
  • AAAI 2027: Additional Reviewer (review completed in 2026)
  • USENIX Security 2027: Reviewer, Cycle 1 (review completed in 2026)

The fuller picture.

Research & teaching experience

Graduate Research Assistant

The University of Texas at Arlington
May 2022 – Present | Arlington, TX

  • Leading research on trustworthy AI systems by adapting software testing, verification, and debugging principles to evaluate and explain CNNs and Vision Transformers
  • Developed DD-CAM, a delta-debugging method for minimal sufficient vision explanations that improved medical image localization by 40% and was accepted as a NeurIPS 2026 oral paper
  • Pioneered ABLE (Adversarial Boundary Local Explanations), using adversarial pairs to construct interpretable local surrogate models – accepted for an oral presentation at KDD 2026
  • Developed combinatorial testing approaches for synthetic data generation using VAEs and GANs, improving data diversity while maintaining distributional fidelity
  • Collaborating with NIST researchers (Dr. Rick Kuhn, Dr. Raghu Kacker) on AI standards and trustworthy ML systems

Graduate Teaching Assistant

The University of Texas at Arlington
Fall 2024 – Present | Arlington, TX

  • Courses: Software Testing & Maintenance (Fall 2024); Introduction to Computers & Programming (Spring 2025); Discrete Structures (Fall 2025); Professional Practices (Spring 2026 and Fall 2026)
  • Prepared assignments, grading materials, and instructional resources across these undergraduate computer science courses
  • Led selected class sessions and hands-on workshops; supported students through office hours, one-on-one assistance, assignment feedback, and project troubleshooting

Guest Lecturer – Introduction to Machine Learning

Winston-Salem State University
November 2025 | Winston-Salem, NC

  • Delivered a 2-week lecture series on Introduction to Machine Learning for undergraduate students
  • Covered foundational ML concepts including supervised/unsupervised learning, neural networks, and practical applications
Industry & entrepreneurship

Co-Founder & CTO

Bhasha Tech, Inc. – Bhasha Language Learning App
January 2025 – Present | Dallas, TX

  • Co-founded and scaled Bhasha, a language-learning platform, to 70,000+ users across 130+ countries
  • Designed an LLM-backed personalization architecture for iOS and Android, solving cross-platform consistency across adaptive lessons and tutoring workflows
  • Integrated speech recognition, pronunciation assessment, and conversational tutoring into an end-to-end ML pipeline that transforms learner speech into personalized feedback

Machine Learning Software Developer Intern

State Farm – Life Fit Project
January 2021 – August 2021 | Richardson, TX

  • Designed an ML pipeline generating health risk scores using K-Means clustering on Fitbit API data for insurance underwriting
  • Improved model accuracy from 78% to 93% by implementing GAN-based data augmentation for imbalanced classes
  • Deployed a scalable backend using AWS (Lambda, SageMaker, EC2, S3, Cognito) serving 10K+ daily predictions
Honors & awards
  • Hesed Endowed Scholarship Fund Award – UTA College of Engineering (2026)
  • Cyneta Networks Outstanding Graduate Teaching Assistant Award – UTA College of Engineering (2026)
  • MavPitch Competition Winner – $40,000 Award (2026)
  • Distinguished Paper Award Candidate – IEEE/ACM CAIN 2024
  • Summa Cum Laude – B.S. Software Engineering, UTA (2021)
Tools & technical skills
  • AI & Deep Learning: PyTorch, TensorFlow, scikit-learn, XGBoost, CNNs, Transformers, Vision Transformers, CLIP, Diffusion Models, Sparse Autoencoders
  • Trustworthy & Explainable AI: Model Debugging and Repair, Adversarial Robustness, Grad-CAM, SHAP, LIME, Concept-Based Methods, Combinatorial Testing
  • Generative & Language AI: GANs, VAEs, Large Language Models (LLMs), LLM APIs, Conversational AI, Speech Recognition, Personalization Systems
  • Programming: Python, C/C++, Java, JavaScript, SQL, Shell Scripting
  • Cloud & MLOps: AWS (SageMaker, Lambda, EC2, S3), Docker, Git, MLflow, Weights & Biases

Let's connect

Good work starts
with a conversation.

Interested in explainable AI, model repair, or building useful ML systems? I'd be glad to hear from you.

Say hello