Publications 
Selected Publications
- CircuitKIT — Pre-Print 2026 · Circuit discovery, evaluation, and application toolkit for mechanistic interpretability
- C-ΔΘ: Circuit-Restricted Weight Arithmetic for Selective Refusal — Mechanistic Interpretability Workshop, ICML 2026 · Circuit-level weight edits that instil selective refusal — no inference-time steering, no runtime overhead
- CuratorKIT — Pre-Print 2026 · Provenance-grounded data-curation and synthetic-generation pipeline for LLM post-training
- SafeTune — Pre-Print 2026 · Unified library for auditing and repairing safety drift in fine-tuned LLMs
- AlignTune — Pre-Print 2026 · One interface for SFT, DPO, GRPO, and RLHF with modular reward framework and interchangeable backends
- Forgetting That Sticks — Pre-Print 2026 · Unlearning that survives quantization by identifying and zeroing the circuits that store the target knowledge
- Interpretability as Alignment — EurIPS Workshop 2025 (Spotlight) · Argues mechanistic interpretability should be a design principle in post-training, not a post-hoc audit
- TabTune — WWW 2026 Demo · Unified library for tabular foundation model inference, fine-tuning, and benchmarking across 7 architectures
- Alberta Wells Dataset — ICML 2025 · Satellite benchmark for detecting abandoned oil & gas wells; climate AI work with Mila / McGill
- SELF-PERCEPT — ACL 2025 · LLM introspection improves detection of multi-person mental manipulation in multi-turn conversations
Safety Post-Training & Alignment
AlignTune: Modular Toolkit for Post-Training Alignment of Large Language Models
R E Zera Marveen Lyngkhoi, Chirag Chawla, Pratinav Seth, Utsav Avaiya, Soham Bhattacharjee, Mykola Khandoga, Rui Yuan, Vinay Kumar Sankarapu
C-ΔΘ: Circuit-Restricted Weight Arithmetic for Selective Refusal
Aditya Kasliwal, Pratinav Seth, Vinay Kumar Sankarapu
-
ALIGNBEAM: Inference-Time Alignment Transfer via Cross-Vocabulary Logit Mixing, Chirag Chawla, Pratinav Seth, Vinay Kumar Sankarapu, AI for Good Workshop, ICML 2026
-
Forgetting That Sticks: Quantization-Permanent Unlearning via Circuit Attribution, Saisab Sadhu, Pratinav Seth, Vinay Kumar Sankarapu, Pre-Print
-
Position: Behavioural Assurance Cannot Verify the Safety Claims Governance Now Demands, Pratinav Seth, Vinay Kumar Sankarapu, Pre-Print
-
What Do Compliance Detectors Read? An Audit of Activation Probes and Guard Models, Saisab Sadhu, Aadit Sengupta, Vinay Kumar Sankarapu, Pratinav Seth, Under Review
-
Drift Then Repair: A Controlled Cross-Paradigm Audit of Safety in Fine-Tuned LLMs, Pratinav Seth, Anshul Kaushal, Saisab Sadhu, Vinay Kumar Sankarapu, Under Review
-
Self-Calibrating Weight-Arithmetic Safety-Drift Repair, Pratinav Seth, Vinay Kumar Sankarapu, Under Review
-
The Off-Switch Failure: When Safety-Repair Evaluation Rewards Model Collapse, Pratinav Seth, Under Review
-
Provenance-Grounded Gating and Adaptive Recovery in Synthetic Post-Training Data Curation, Soham Bhattacharjee, Karun Sharma, Vinay Kumar Sankarapu, Pratinav Seth, Under Review
-
Document-as-Function: Verifiable Generation of Long-Form Synthetic Documents, Karun Sharma, Soham Bhattacharjee, Vinay Kumar Sankarapu, Pratinav Seth, Under Review
Mechanistic Interpretability & XAI
Interpretability as Alignment: Making Internal Understanding a Design Principle
Aadit Sengupta, Pratinav Seth, Vinay Kumar Sankarapu
Interpretability-aware pruning for efficient medical image analysis
Nikita Malik, Pratinav Seth, Neeraj Kumar Singh, Chintan Chitroda, Vinay Kumar Sankarapu
Bridging the gap in XAI-why reliable metrics matter for explainability and compliance
Pratinav Seth, Vinay Kumar Sankarapu
DLBacktrace: A Model Agnostic Explainability for any Deep Learning Models
Vinay Kumar Sankarapu, Chintan Chitroda, Yashwardhan Rathore, Neeraj Kumar Singh, Pratinav Seth
-
xai_evals: A Framework for Evaluating Post-Hoc Local Explanation Methods,Pratinav Seth, Yashwardhan Rathore, Neeraj Kumar Singh, Chintan Chitroda, Vinay Kumar Sankarapu, Technical Report Citations
-
CircuitKIT: Circuit Discovery, Evaluation, and Application Toolkit for Mechanistic Interpretability, Pratinav Seth, Hem Gosalia, Aditya Kasliwal, Vinay Kumar Sankarapu, Pre-Print
-
Faithfulness to Refusal: A Causal Audit of Neuron Selectors in LLMs, Ananth Eswar, Pratinav Seth, Utsav Avaiya, Vinay Kumar Sankarapu, Under Review
-
Faithfulness Is Not Actionability: Component Heterogeneity in Discovered Circuits, Pratinav Seth, Hem Gosalia, Aditya Kasliwal, Vinay Kumar Sankarapu, Under Review
-
DLBacktrace v2: Extending Model-Agnostic Interpretability for LLMs and MoEs with CUDA Acceleration, Neeraj Kumar Singh, Pratinav Seth, Omkar Kakade, Chintan Chitroda, Vinay Kumar Sankarapu, Pre-Print
Tabular Foundation Models & Distillation
Orion-MSP: Multi-Scale Sparse Attention for Tabular In-Context Learning
Mohamed Bouadi, Pratinav Seth, Aditya Tanna, Vinay Kumar Sankarapu
Orion-BiX: Bi-Axial Attention for Tabular In-Context Learning
Mohamed Bouadi, Pratinav Seth, Aditya Tanna, Vinay Kumar Sankarapu
Exploring Fine-Tuning for Tabular Foundation Models
Aditya Tanna, Pratinav Seth, Mohamed Bouadi, Vinay Kumar Sankarapu
TabTune: A Unified Library for Inference and Fine-Tuning Tabular Foundation Models (Demo)
Aditya Tanna, Pratinav Seth, Mohamed Bouadi, Utsav Avaiya, Vinay Kumar Sankarapu
-
Pocket Foundation Models: Distilling TFMs into CPU-Ready Gradient-Boosted Trees, Aditya Tanna, Nassim Bouarour, Mohamed Bouadi, Vinay Kumar Sankarapu, Pratinav Seth, Foundation Models for Structured Data (FMSD) Workshop, ICML 2026
-
Ensembling Tabular Foundation Models: A Diversity Ceiling and a Calibration Trap, Aditya Tanna, Yash Jignesh Desai, Pratinav Seth, Mohamed Bouadi, Nassim Bouarour, Vinay Kumar Sankarapu, Foundation Models for Structured Data (FMSD) Workshop, ICML 2026
-
Distilling Tabular Foundation Models for Structured Health Data, Aditya Tanna, Nassim Bouarour, Mohamed Bouadi, Vinay Kumar Sankarapu, Pratinav Seth, Structured Data for Health (SD4H) Workshop, ICML 2026 — Best Paper Runner-Up, Spotlight
-
Data Presentation over Architecture: Resampling Strategies for Credit Risk Prediction with Tabular Foundation Models, Aditya Tanna, Mitul Solanki, Mohamed Bouadi, Nassim Bouarour, Pratinav Seth, Vinay Kumar Sankarapu, FinDS Workshop @ ACM SIGMOD 2026 (Oral)
Libraries & Toolkits
Designed and built the initial release of each library below.
- TabTune — Unified inference and fine-tuning library for tabular foundation models across 7 architectures. WWW 2026 Demo (116⭐).
- AlignTune — Modular post-training toolkit: SFT, DPO, GRPO, and RLHF with interchangeable backends. Pre-Print 2026 (37⭐).
- DLBacktrace v2 —
torch.export-based, model-agnostic explainability for LLMs and MoEs with CUDA acceleration. Pre-Print 2026 (26⭐). - CuratorKIT — Provenance-grounded data-curation and synthetic-generation pipeline for LLM post-training. Pre-Print 2026 (24⭐).
- xai_evals — Framework for evaluating post-hoc local explanation methods. Technical Report 2025 (15⭐).
- CircuitKIT — Circuit discovery, evaluation, and application toolkit for mechanistic interpretability. Pre-Print 2026 (14⭐).
- SafeTune — Unified library for auditing and repairing safety drift in fine-tuned LLMs. Pre-Print 2026 (7⭐).
- DLBacktrace (v1) — Model-agnostic explainability for deep learning models. IJCNN 2025.
In development: multiple additional tools across interpretability, post-training alignment, and LLM safety.
Climate Change & Earth Observation
Alberta Wells Dataset: Pinpointing Oil and Gas Wells from Satellite Imagery
Pratinav Seth(#), Michelle Lin(#), Brefo Dwamena Yaw, Jade Boutot, Mary Kang, David Rolnick
-
Performance Evaluation of Deep Segmentation Models for Contrails Detection, Akshat Bhandari, Sriya Rallabandi, Sanchit Singhal, Aditya Kasliwal, Pratinav Seth, Tackling Climate Change with Machine Learning Workshop at NeurIPS 2022. Citations
-
Sailing Through Spectra: Unveiling the Potential of Multi-Spectral Information in Marine Debris Segmentation, Dyutit Mohanty, Aditya Kasliwal, Bharath Udapa, Pratinav Seth, The Second Tiny Papers Track at ICLR 2024. Citations
Medical Imaging
Obscure to Observe: A Lesion-Aware MAE for Glaucoma Detection from Retinal Context
Siddhant Bharadwaj, Pratinav Seth, Chandra Sekhar Seelamantula
-
ReFuSeg: Regularized Multi-Modal Fusion for Precise Brain Tumour Segmentation, Aditya Kasliwal, Sankarshanaa Sagaram, Laven Srivastava, Pratinav Seth, Adil Khan, 9th Edition of the Brain Lesion (BrainLes) workshop, MICCAI 2023. Citations
-
UATTA-ENS: Uncertainty Aware Test Time Augmented Ensemble for PIRC Diabetic Retinopathy Detection, Pratinav Seth, Adil Khan, Ananya Gupta, Saurabh Kumar Mishra, Akshat Bhandhari, Medical Imaging meets NeurIPS Workshop, NeurIPS 2022. Citations
Uncertainty & Robustness
-
Evaluating Predictive Uncertainty and Robustness to Distributional Shift Using Real World Data, Kumud Lakara (†), Akshat Bhandari (†), Pratinav Seth (†), Ujjwal Verma, Bayesian Deep Learning Workshop, NeurIPS 2021. Citations
-
UATTA-EB: Uncertainty-Aware Test-Time Augmented Ensemble of BERTs for Classifying Common Mental Illnesses on Social Media Posts, Pratinav Seth (†), Mihir Agarwal (†), 1st Tiny Paper Track at ICLR 2023. Citations
-
Analyzing Effects of Fake Training Data on the Performance of Deep Learning Systems, Pratinav Seth (†), Akshat Bhandari (†), Kumud Lakara (†), Pre-Print Citations
NLP & AI for Social Good
Danush Khanna, Pratinav Seth, Sidhaarth Sredharan Murali, Aditya Kumar Guru, Siddharth Shukla, Tanuj Tyagi, Sandeep Chaurasia, Kripabandhu Ghosh
AgriLLM: Harnessing transformers for farmer queries.
Krish Didwania (†), Pratinav Seth (†), Aditya Kasliwal, Amit Agarwal
-
SSS at SemEval-2023 Task 10: Explainable Detection of Online Sexism using Majority Voted Fine-Tuned Transformers, Sriya Rallabandi, Sanchit Singhal, Pratinav Seth, Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023), ACL 2023 Citations
-
RSM-NLP at BLP-2023 Task 2: Bangla Sentiment Analysis using Weighted and Majority Voted Fine-Tuned Transformers, Pratinav Seth, Rashi Goel, Komal Mathur, Swetha Vemulapalli, Proceedings of the 1st Workshop on Bangla Language Processing (BLP 2023), EMNLP 2023 Citations
-
HGP-NLP at Shared Task: Leveraging LoRA for Lay Summarization of Biomedical Research Articles using Seq2Seq Transformers, Hemang Malik, Gaurav Pradeep, Pratinav Seth, Accepted at BioNLP 2024 Workshop, ACL 2024. Citations
Vision & Super-Resolution
CoReFusion: Contrastive Regularized Fusion for Guided Thermal Super-Resolution
Aditya Kasliwal, Pratinav Seth, Sriya Rallabandi, Sanchit Singhal
LaMAR: Laplacian Pyramid for Multimodal Adaptive Super Resolution (Student Abstract)
Aditya Kasliwal, Aryan Kamani, Ishaan Gakhar, Pratinav Seth, Sriya Rallabandi
Laplacian reconstructive network for guided thermal super-resolution
Aditya Kasliwal, Ishaan Gakhar, Aryan Kamani, Pratinav Seth, Ujjwal Verma
Selected Blog Posts
- Retail Banking Case Study: Why Specialists Beat Generalists on Template-Strict Workflows, Chirag Chawla, Zera Lyngkhoi, Pratinav Seth, Utsav Avaiya, Soham Bhattacharjee, Mykola Khandoga, Rui Yuan, Vinay Kumar Sankarapu, Lexsi Labs, 2026
- The Specialization Dividend: Aligning a 4B Model for Wealth Management Using AlignTune, Chirag Chawla, Zera Lyngkhoi, Pratinav Seth, Utsav Avaiya, Soham Bhattacharjee, Mykola Khandoga, Rui Yuan, Vinay Kumar Sankarapu, Lexsi Labs, 2026
- Now Shipping: TabTune Regression for Tabular Foundation Models, Aditya Tanna, Pratinav Seth, Mohamed Bouadi, Utsav Avaiya, Vinay Kumar Sankarapu, Lexsi Labs, 2026