Aniket Ghosh.

Selected for A Diagnostic Panel of Ground-Truth Probes for Language Models, mentored by Mohamed Amine Merzouk (postdoctoral researcher, Mila & McGill) and Adam Oberman (full professor at McGill, member of Mila, research advisor at LawZero). Most safety probing starts from a human concept — deception, sandbagging, sycophancy — and hunts for its direction in activation space, inheriting noisy labels and fragile transfer. This project inverts that: it admits only probe targets whose ground truth is exact and computable from the output itself, so every claim stays falsifiable.

• Training ground-truth probes: end-of-sequence hazard, repetition onset, copy-versus-generate, format validity
• Scored on calibration rather than noisy human labels, giving the method a validation floor
• Building outward from the team's length-probe work toward a joint diagnostic panel

TA for CS 5800 Algorithms under Professor Luowen Qian, covering algorithm design and analysis for the graduate core.

• Supporting graduate students across problem sets, office hours, and grading
• Second TA appointment at Khoury, after the graduate NLP course

The only person on Varosync's ML stack, building it from scratch and reporting directly to the CEO and CTO. I owned it end to end: the knowledge graph, the molecular embedding models, and the pipeline through to the live query path.

• Awarded the Summer 2026 Research Fellowship (Varosync, Inc. with NYCEDC's NYC AI Nexus) for delivering Trialfecta's v0 asset evidence pipeline, safety-aware drug embeddings and benchmark evaluation harness · View credential
• Built a biomedical knowledge graph from scratch, spanning millions of molecules
• Trained 7–8B parameter molecular embedding models on an 800M-molecule corpus across a multi-GPU cluster, owning data curation, distributed training, checkpointing and evaluation
• Caught the retrieval benchmark training on future outcomes in 123 of 133 cases; swapped k-fold for a leave-one-group-out, point-in-time split and took post-cutoff training evidence from 21.8% to 0%
• Shipped my published probe-guided quantization method into production, deciding where capacity goes across layers in those models. Paper in preparation.
• Owned everything downstream of the gold labels: featurization, vector indexing, retrieval logic, and the user query path
• Owned ML infrastructure across a hybrid cloud stack: AWS, Nebius (GPU training), and Mithril (inference/batch)
• Claude Code and Cursor wired into the daily agentic loop

Nine-week summer intensive led by Kathleen (Katie) Creel, funded by the Alfred P. Sloan Foundation, bringing together ~10-12 philosophy PhD students from institutions across the country for training in ML, AI ethics, and metascience. Designed and led a 12-session hands-on ML lab built from scratch: Python, fairness, differential privacy, neural networks by hand, transformers, epistemic simulations, and mechanistic interpretability.

• Designed and led 12 hands-on ML lab sessions for philosophy PhD students
• Taught sparse autoencoders, the logit lens, and linear probing hands-on using nnsight and NDIF
• Built curriculum from the ground up on fairness, interpretability, and agent-based modeling
• Program funded by the Alfred P. Sloan Foundation with a metascience focus

Taught the PyTorch and deep-learning labs for Professor Amir Tahmasebi's graduate NLP course, from fundamentals through transformer architectures.

• Ran weekly hands-on labs: PyTorch basics through transformer implementation
• Supported 40+ graduate students across assignments, projects, and grading

Developed custom ML pipelines for cell segmentation in histopathological images under Dr. Kaushiki Roy and Dr. Debotosh Bhattacharjee. Implemented preprocessing techniques using color quantization and DBSCAN clustering for densely packed cell detection. Guided junior researchers in deep learning for medical imaging.

• Built ML pipelines combining classical CV and deep learning approaches
• Implemented color quantization and DBSCAN clustering for cell detection
• Guided junior researchers in deep learning for medical imaging

Contributed to developing an Indian Sign Language dataset under Dr. Arnab Kumar Maji. Defined quality criteria and annotation standards by analyzing existing datasets including Microsoft ASL, MNIST, and Static ISL to establish best practices.

Facilitated IT system migration from local servers to Oracle cloud infrastructure. Supported IT operations for a 1000+ room property.

Taught mathematics and computer science to 10 underprivileged students over seven years. Developed custom learning materials and structured lesson plans that broke complex topics into manageable concepts.

• 7 years of active teaching to underprivileged students
• Developed custom learning materials for math and CS
• Provided mentorship beyond academic subjects