Education & service

• 2025.12Selected as Mentor for the AAAI Undergraduate Consortium 2026, returning after being a Scholar in the 2023 cohort.
• 2023.02One of 11 undergraduates selected as an AAAI Undergraduate Consortium Scholar 2023, with a $2,000 travel grant to present at AAAI-23 in Washington D.C.
• 2023.01MAHE Undergraduate Research Grant (₹10,000) for Explainable & Trustworthy Skin Lesion Classification, under Dr. Abhilash K. Pai.
• 2022.06Top 10 team of 1,000+ submissions, Bajaj Finserv HackRx 3.0 Hackathon.

• MentoringMentor, AAAI Undergraduate Consortium 2026 (Scholar 2023 → Mentor 2026)
• ReviewingMain conferences: AAAI 2027 (Program Committee), NeurIPS 2026, ACM AIES 2026, CVPR (2025–26), ECCV (2024–26), ICCV 2025, WACV (2026–27), BMVC 2026, IJCNN 2025

Certifications

• Deep Learning Specialization, DeepLearning.AI
• 6th Summer School on AI, CVIT IIIT Hyderabad

• Vinay Kumar Sankarapu32 papersManagerLexsi Labs / AryaXAI Alignment Labs

  • AlignTune: Modular Toolkit for Post-Training Alignment of Large Language Models
  • C-ΔΘ: Circuit-Restricted Weight Arithmetic for Selective Refusal
  • ALIGNBEAM: Inference-Time Alignment Transfer via Cross-Vocabulary Logit Mixing
  • Forgetting That Sticks: Quantization-Permanent Unlearning via Circuit Attribution
  • Position: Behavioural Assurance Cannot Verify the Safety Claims Governance Now Demands
  • What Do Compliance Detectors Read? An Audit of Activation Probes and Guard Models
  • Verdict Without the Rule: Diagnosing and Auditing Regulatory Rule Sensitivity in LLM Compliance Systems
  • Every Ablation Is a Dose: Counterweights and the Semblance of Self-Repair
  • AgentTune: A Unified Training Lifecycle for Tool-Using LLM Agents
  • SafeTune: A Unified Faithful Library for Auditing and Repairing Safety Drift in Fine-Tuned LLMs
  • Drift, Then Repair: A Safety Audit of Fine-Tuned Language Models
  • Self-Calibrating Weight-Arithmetic Safety-Drift Repair
  • CuratorKIT: Data Curation and Synthetic Data Generation for LLM Post-Training
  • Provenance-Grounded Gating and Adaptive Recovery in Synthetic Post-Training Data Curation
  • Document-as-Function: Verifiable Generation of Long-Form Synthetic Documents
  • Interpretability as Alignment: Making Internal Understanding a Design Principle
  • Interpretability-aware pruning for efficient medical image analysis
  • Bridging the gap in XAI-why reliable metrics matter for explainability and compliance
  • DLBacktrace: A Model Agnostic Explainability for any Deep Learning Models
  • xai_evals: A Framework for Evaluating Post-Hoc Local Explanation Methods
  • CircuitKIT: Circuit Discovery, Evaluation, and Application Toolkit for Mechanistic Interpretability
  • Faithfulness to Refusal: A Causal Audit of Neuron Selectors in LLMs
  • Faithfulness Is Not Actionability: Extrinsic Validation of Circuit Selectors Under Structured Pruning
  • DLBacktrace v2: Extending Model-Agnostic Interpretability for LLMs and MoEs with CUDA Acceleration
  • Orion-MSP: Multi-Scale Sparse Attention for Tabular In-Context Learning
  • Orion-BiX: Bi-Axial Attention for Tabular In-Context Learning
  • Exploring Fine-Tuning for Tabular Foundation Models
  • TabTune: A Unified Library for Inference and Fine-Tuning Tabular Foundation Models (Demo)
  • Pocket Foundation Models: Distilling TFMs into CPU-Ready Gradient-Boosted Trees
  • Ensembling Tabular Foundation Models: A Diversity Ceiling and a Calibration Trap
  • Distilling Tabular Foundation Models for Structured Health Data
  • Data Presentation over Architecture: Resampling Strategies for Credit Risk Prediction with Tabular Foundation Models
• Aditya Kasliwal10 papersMenteeMars Rover Manipal / Research Society Manipal / Lexsi Labs

  • C-ΔΘ: Circuit-Restricted Weight Arithmetic for Selective Refusal
  • CircuitKIT: Circuit Discovery, Evaluation, and Application Toolkit for Mechanistic Interpretability
  • Faithfulness Is Not Actionability: Extrinsic Validation of Circuit Selectors Under Structured Pruning
  • Performance Evaluation of Deep Segmentation Models for Contrails Detection
  • Sailing Through Spectra: Unveiling the Potential of Multi-Spectral Information in Marine Debris Segmentation
  • ReFuSeg: Regularized Multi-Modal Fusion for Precise Brain Tumour Segmentation
  • AgriLLM: Harnessing transformers for farmer queries.
  • CoReFusion: Contrastive Regularized Fusion for Guided Thermal Super-Resolution
  • LaMAR: Laplacian Pyramid for Multimodal Adaptive Super Resolution (Student Abstract)
  • Laplacian reconstructive network for guided thermal super-resolution
• Aditya Tanna8 papersMenteeLexsi Labs

  • Orion-MSP: Multi-Scale Sparse Attention for Tabular In-Context Learning
  • Orion-BiX: Bi-Axial Attention for Tabular In-Context Learning
  • Exploring Fine-Tuning for Tabular Foundation Models
  • TabTune: A Unified Library for Inference and Fine-Tuning Tabular Foundation Models (Demo)
  • Pocket Foundation Models: Distilling TFMs into CPU-Ready Gradient-Boosted Trees
  • Ensembling Tabular Foundation Models: A Diversity Ceiling and a Calibration Trap
  • Distilling Tabular Foundation Models for Structured Health Data
  • Data Presentation over Architecture: Resampling Strategies for Credit Risk Prediction with Tabular Foundation Models
• Mohamed Bouadi8 papersLexsi Labs, Paris

  • Orion-MSP: Multi-Scale Sparse Attention for Tabular In-Context Learning
  • Orion-BiX: Bi-Axial Attention for Tabular In-Context Learning
  • Exploring Fine-Tuning for Tabular Foundation Models
  • TabTune: A Unified Library for Inference and Fine-Tuning Tabular Foundation Models (Demo)
  • Pocket Foundation Models: Distilling TFMs into CPU-Ready Gradient-Boosted Trees
  • Ensembling Tabular Foundation Models: A Diversity Ceiling and a Calibration Trap
  • Distilling Tabular Foundation Models for Structured Health Data
  • Data Presentation over Architecture: Resampling Strategies for Credit Risk Prediction with Tabular Foundation Models
• Saisab Sadhu6 papersMenteeLexsi Labs

  • Forgetting That Sticks: Quantization-Permanent Unlearning via Circuit Attribution
  • What Do Compliance Detectors Read? An Audit of Activation Probes and Guard Models
  • Cited but Not Consulted: A Counterfactual Audit of Legal Chain-of-Thought Faithfulness
  • Verdict Without the Rule: Diagnosing and Auditing Regulatory Rule Sensitivity in LLM Compliance Systems
  • SafeTune: A Unified Faithful Library for Auditing and Repairing Safety Drift in Fine-Tuned LLMs
  • Drift, Then Repair: A Safety Audit of Fine-Tuned Language Models
• Soham Bhattacharjee5 papersMenteeLexsi Labs

  • AlignTune: Modular Toolkit for Post-Training Alignment of Large Language Models
  • CuratorKIT: Data Curation and Synthetic Data Generation for LLM Post-Training
  • Provenance-Grounded Gating and Adaptive Recovery in Synthetic Post-Training Data Curation
  • Document-as-Function: Verifiable Generation of Long-Form Synthetic Documents
  • BRIEF: Bounded-Context Regeneration of Long Documents with Verifiable Structural Topology
• Akshat Bhandari4 papersMars Rover Manipal

  • Performance Evaluation of Deep Segmentation Models for Contrails Detection
  • UATTA-ENS: Uncertainty Aware Test Time Augmented Ensemble for PIRC Diabetic Retinopathy Detection
  • Evaluating Predictive Uncertainty and Robustness to Distributional Shift Using Real World Data
  • Analyzing Effects of Fake Training Data on the Performance of Deep Learning Systems
• Chintan Chitroda4 papersManagerAryaXAI Alignment Labs

  • Interpretability-aware pruning for efficient medical image analysis
  • DLBacktrace: A Model Agnostic Explainability for any Deep Learning Models
  • xai_evals: A Framework for Evaluating Post-Hoc Local Explanation Methods
  • DLBacktrace v2: Extending Model-Agnostic Interpretability for LLMs and MoEs with CUDA Acceleration
• Nassim Bouarour4 papersLexsi Labs, Paris

  • Pocket Foundation Models: Distilling TFMs into CPU-Ready Gradient-Boosted Trees
  • Ensembling Tabular Foundation Models: A Diversity Ceiling and a Calibration Trap
  • Distilling Tabular Foundation Models for Structured Health Data
  • Data Presentation over Architecture: Resampling Strategies for Credit Risk Prediction with Tabular Foundation Models
• Neeraj Kumar Singh4 papersAryaXAI Alignment Labs

  • Interpretability-aware pruning for efficient medical image analysis
  • DLBacktrace: A Model Agnostic Explainability for any Deep Learning Models
  • xai_evals: A Framework for Evaluating Post-Hoc Local Explanation Methods
  • DLBacktrace v2: Extending Model-Agnostic Interpretability for LLMs and MoEs with CUDA Acceleration
• Sriya Rallabandi4 papersMars Rover Manipal

  • Performance Evaluation of Deep Segmentation Models for Contrails Detection
  • SSS at SemEval-2023 Task 10: Explainable Detection of Online Sexism using Majority Voted Fine-Tuned Transformers
  • CoReFusion: Contrastive Regularized Fusion for Guided Thermal Super-Resolution
  • LaMAR: Laplacian Pyramid for Multimodal Adaptive Super Resolution (Student Abstract)
• Aadit Sengupta3 papersMenteeLexsi Labs

  • What Do Compliance Detectors Read? An Audit of Activation Probes and Guard Models
  • Verdict Without the Rule: Diagnosing and Auditing Regulatory Rule Sensitivity in LLM Compliance Systems
  • Interpretability as Alignment: Making Internal Understanding a Design Principle
• Karun Sharma3 papersMenteeLexsi Labs

  • CuratorKIT: Data Curation and Synthetic Data Generation for LLM Post-Training
  • Provenance-Grounded Gating and Adaptive Recovery in Synthetic Post-Training Data Curation
  • Document-as-Function: Verifiable Generation of Long-Form Synthetic Documents
• Sanchit Singhal3 papersMars Rover Manipal

  • Performance Evaluation of Deep Segmentation Models for Contrails Detection
  • SSS at SemEval-2023 Task 10: Explainable Detection of Online Sexism using Majority Voted Fine-Tuned Transformers
  • CoReFusion: Contrastive Regularized Fusion for Guided Thermal Super-Resolution
• Utsav Avaiya3 papersMenteeLexsi Labs

  • AlignTune: Modular Toolkit for Post-Training Alignment of Large Language Models
  • Faithfulness to Refusal: A Causal Audit of Neuron Selectors in LLMs
  • TabTune: A Unified Library for Inference and Fine-Tuning Tabular Foundation Models (Demo)
• Adil Khan2 papersMars Rover Manipal

  • ReFuSeg: Regularized Multi-Modal Fusion for Precise Brain Tumour Segmentation
  • UATTA-ENS: Uncertainty Aware Test Time Augmented Ensemble for PIRC Diabetic Retinopathy Detection
• Anshul Kaushal2 papersMenteeLexsi Labs

  • SafeTune: A Unified Faithful Library for Auditing and Repairing Safety Drift in Fine-Tuned LLMs
  • Drift, Then Repair: A Safety Audit of Fine-Tuned Language Models
• Aryan Kamani2 papersMars Rover Manipal

  • LaMAR: Laplacian Pyramid for Multimodal Adaptive Super Resolution (Student Abstract)
  • Laplacian reconstructive network for guided thermal super-resolution
• Chirag Chawla2 papersMenteeLexsi Labs

  • AlignTune: Modular Toolkit for Post-Training Alignment of Large Language Models
  • ALIGNBEAM: Inference-Time Alignment Transfer via Cross-Vocabulary Logit Mixing
• Hem Gosalia2 papersMenteeLexsi Labs

  • CircuitKIT: Circuit Discovery, Evaluation, and Application Toolkit for Mechanistic Interpretability
  • Faithfulness Is Not Actionability: Extrinsic Validation of Circuit Selectors Under Structured Pruning
• Ishaan Gakhar2 papersMars Rover Manipal

  • LaMAR: Laplacian Pyramid for Multimodal Adaptive Super Resolution (Student Abstract)
  • Laplacian reconstructive network for guided thermal super-resolution
• Kumud Lakara2 papersMars Rover Manipal

  • Evaluating Predictive Uncertainty and Robustness to Distributional Shift Using Real World Data
  • Analyzing Effects of Fake Training Data on the Performance of Deep Learning Systems
• R E Zera Marveen Lyngkhoi2 papersMenteeLexsi Labs

  • AlignTune: Modular Toolkit for Post-Training Alignment of Large Language Models
  • AgentTune: A Unified Training Lifecycle for Tool-Using LLM Agents
• Ujjwal Verma2 papersMentorMIT MAHE

  • Evaluating Predictive Uncertainty and Robustness to Distributional Shift Using Real World Data
  • Laplacian reconstructive network for guided thermal super-resolution
• Yashwardhan Rathore2 papersAryaXAI Alignment Labs

  • DLBacktrace: A Model Agnostic Explainability for any Deep Learning Models
  • xai_evals: A Framework for Evaluating Post-Hoc Local Explanation Methods
• Abhilash K Pai1 paperMentorMIT MAHE

  • Does the Fairness of Your Pre-Training Hold Up? Examining the Influence of Pre-Training Techniques on Skin Tone Bias in Skin Lesion Classification
• Abhivansh Gupta1 paperLexsi Labs

  • AgentTune: A Unified Training Lifecycle for Tool-Using LLM Agents
• Aditya Kumar Guru1 paperManipal University Jaipur

  • SELF-PERCEPT: Introspection Improves Large Language Models' Detection of Multi-Person Mental Manipulation in Conversations
• Amit Agarwal1 paperWells Fargo AI COE

  • AgriLLM: Harnessing transformers for farmer queries.
• Ananth Eswar1 paperMenteeLexsi Labs

  • Faithfulness to Refusal: A Causal Audit of Neuron Selectors in LLMs
• Ananya Gupta1 paperMars Rover Manipal

  • UATTA-ENS: Uncertainty Aware Test Time Augmented Ensemble for PIRC Diabetic Retinopathy Detection
• Areeb Ahmad1 paperLexsi Labs

  • Every Ablation Is a Dose: Counterweights and the Semblance of Self-Repair
• Aryan Sood1 paperLexsi Labs

  • BRIEF: Bounded-Context Regeneration of Long Documents with Verifiable Structural Topology
• Bharath Udapa1 paperResearch Society Manipal

  • Sailing Through Spectra: Unveiling the Potential of Multi-Spectral Information in Marine Debris Segmentation
• Brefo Dwamena Yaw1 paperMcGill University

  • Alberta Wells Dataset: Pinpointing Oil and Gas Wells from Satellite Imagery
• Chandra Sekhar Seelamantula1 paperMIT Manipal / IISc Bangalore

  • Obscure to Observe: A Lesion-Aware MAE for Glaucoma Detection from Retinal Context
• Danush Khanna1 paperManipal University Jaipur

  • SELF-PERCEPT: Introspection Improves Large Language Models' Detection of Multi-Person Mental Manipulation in Conversations
• David Rolnick1 paperMentorMila / McGill

  • Alberta Wells Dataset: Pinpointing Oil and Gas Wells from Satellite Imagery
• Dyutit Mohanty1 paperResearch Society Manipal

  • Sailing Through Spectra: Unveiling the Potential of Multi-Spectral Information in Marine Debris Segmentation
• Gaurav Pradeep1 paperResearch Society Manipal

  • HGP-NLP at Shared Task: Leveraging LoRA for Lay Summarization of Biomedical Research Articles using Seq2Seq Transformers
• Hemang Malik1 paperResearch Society Manipal

  • HGP-NLP at Shared Task: Leveraging LoRA for Lay Summarization of Biomedical Research Articles using Seq2Seq Transformers
• Jade Boutot1 paperMcGill University

  • Alberta Wells Dataset: Pinpointing Oil and Gas Wells from Satellite Imagery
• Komal Mathur1 paperResearch Society Manipal

  • RSM-NLP at BLP-2023 Task 2: Bangla Sentiment Analysis using Weighted and Majority Voted Fine-Tuned Transformers
• Kripabandhu Ghosh1 paperIISER Kolkata

  • SELF-PERCEPT: Introspection Improves Large Language Models' Detection of Multi-Person Mental Manipulation in Conversations
• Krish Didwania1 paperMars Rover Manipal

  • AgriLLM: Harnessing transformers for farmer queries.
• Laven Srivastava1 paperMars Rover Manipal

  • ReFuSeg: Regularized Multi-Modal Fusion for Precise Brain Tumour Segmentation
• Mary Kang1 paperMcGill University

  • Alberta Wells Dataset: Pinpointing Oil and Gas Wells from Satellite Imagery
• Michelle Lin1 paperMcGill University

  • Alberta Wells Dataset: Pinpointing Oil and Gas Wells from Satellite Imagery
• Mihir Agarwal1 paperResearch Society Manipal

  • UATTA-EB: Uncertainty-Aware Test-Time Augmented Ensemble of BERTs for Classifying Common Mental Illnesses on Social Media Posts
• Mitul Solanki1 paperMenteeLexsi Labs

  • Data Presentation over Architecture: Resampling Strategies for Credit Risk Prediction with Tabular Foundation Models
• Mykola Khandoga1 paperLexsi Labs, Paris

  • AlignTune: Modular Toolkit for Post-Training Alignment of Large Language Models
• Nikita Malik1 paperMenteeLexsi Labs

  • Interpretability-aware pruning for efficient medical image analysis
• Omkar Kakade1 paperMenteeLexsi Labs

  • DLBacktrace v2: Extending Model-Agnostic Interpretability for LLMs and MoEs with CUDA Acceleration
• Ram Mohan Rao Kadiyala1 paperLexsi Labs

  • AgentTune: A Unified Training Lifecycle for Tool-Using LLM Agents
• Rashi Goel1 paperResearch Society Manipal

  • RSM-NLP at BLP-2023 Task 2: Bangla Sentiment Analysis using Weighted and Majority Voted Fine-Tuned Transformers
• Rui Yuan1 paperLexsi Labs, Paris

  • AlignTune: Modular Toolkit for Post-Training Alignment of Large Language Models
• Sandeep Chaurasia1 paperManipal University Jaipur

  • SELF-PERCEPT: Introspection Improves Large Language Models' Detection of Multi-Person Mental Manipulation in Conversations
• Sankarshanaa Sagaram1 paperMars Rover Manipal

  • ReFuSeg: Regularized Multi-Modal Fusion for Precise Brain Tumour Segmentation
• Saurabh Kumar Mishra1 paperMars Rover Manipal

  • UATTA-ENS: Uncertainty Aware Test Time Augmented Ensemble for PIRC Diabetic Retinopathy Detection
• Shreeyans Arora1 paperLexsi Labs

  • Cited but Not Consulted: A Counterfactual Audit of Legal Chain-of-Thought Faithfulness
• Siddhant Bharadwaj1 paperMIT Manipal / IISc Bangalore

  • Obscure to Observe: A Lesion-Aware MAE for Glaucoma Detection from Retinal Context
• Siddharth Shukla1 paperManipal University Jaipur

  • SELF-PERCEPT: Introspection Improves Large Language Models' Detection of Multi-Person Mental Manipulation in Conversations
• Sidhaarth Sredharan Murali1 paperNIT Surathkal

  • SELF-PERCEPT: Introspection Improves Large Language Models' Detection of Multi-Person Mental Manipulation in Conversations
• Swetha Vemulapalli1 paperResearch Society Manipal

  • RSM-NLP at BLP-2023 Task 2: Bangla Sentiment Analysis using Weighted and Majority Voted Fine-Tuned Transformers
• Tanuj Tyagi1 paperManipal University Jaipur

  • SELF-PERCEPT: Introspection Improves Large Language Models' Detection of Multi-Person Mental Manipulation in Conversations
• Vidushee Vats1 paperLexsi Labs

  • AgentTune: A Unified Training Lifecycle for Tool-Using LLM Agents
• Yash Jignesh Desai1 paperLexsi Labs, Paris

  • Ensembling Tabular Foundation Models: A Diversity Ceiling and a Calibration Trap