Aditya Oke

Carnegie Mellon University

Pittsburgh, PA 15213

I’m a master’s student in Computational Data Science at Carnegie Mellon University, working with Prof. Lei Li in the Li Lab. My research focuses on AI safety, red teaming, and coding agents.

I work on AI safety and evaluation, with prior work on AI-generated content detection, hallucination detection, and safeguards for language models. My current project aligns language models for idiomatic code generation.

I’m also participating in the Amazon Nova Challenge, building red-teaming LLM agents to detect vulnerable code generation and creating safety benchmarks for adversarial coding prompts.

Before CMU, I was a Senior Software Engineer at JPMorgan Chase, where I designed and deployed distributed Spark pipelines for multi-terabyte financial datasets and built AWS-based infrastructure for AI-driven compliance tools.

news

May 08, 2026	Sem-Detect was accepted to ICML 2026.
Feb 01, 2026	Participating in the Amazon Nova Challenge, building red-teaming agents for LLM code generation safety!
Sep 01, 2025	Joined Li Lab as a Graduate Researcher working on LLM safety and robustness!
Aug 25, 2025	Started my MS in Computational Data Science at Carnegie Mellon University!
Jan 10, 2022	Started as Senior Software Engineer at JPMorgan Chase, working on distributed systems and Big Data!

Selected Publications

ICML
Sem-Detect: Semantic Level Detection of AI Generated Peer-Reviews

André V. Duarte, Brian Tufts, Aditya Oke, and 3 more authors

In The 43rd International Conference on Machine Learning (ICML), Seoul, South Korea, Jul 2026

Abs Bib

How can we distinguish whether a peer review was written by a human or generated by an AI model? We argue that, in this setting, authorship should not be attributed solely from the textual features of a review, but also from the ideas, judgments, and claims it expresses. To this end, we propose Sem-Detect, an authorship detection method for peer reviews that operationalizes this principle by combining textual features with claim-level semantic analysis. Sem-Detect compares a target review against multiple AI-generated reviews of the same paper, leveraging the observation that different AI models tend to converge on similar points, while human reviewers introduce more unique and diverse ones. As a result, Sem-Detect is able to distinguish fully AI-generated reviews from authentic human-written ones, including those that have been refined using an LLM but still reflect human judgment. Across a dataset of over 20,000 peer reviews from ICLR and NeurIPS conferences, Sem-Detect improves over the strongest prior detector by 36.5% in TPR@1% FPR in the binary setting. More importantly, in the three-class scenario, we empirically show that LLM refinement preserves the semantic signals of human reviews, which remain distinct from the patterns exhibited by fully AI-generated text; as a result, fewer than 3.5% of LLM-refined human reviews are misclassified as AI-generated.
@inproceedings{duarte2026sem, title = {Sem-Detect: Semantic Level Detection of {AI} Generated Peer-Reviews}, author = {Duarte, Andr{\'e} V. and Tufts, Brian and Oke, Aditya and Fang, Fei and Oliveira, Arlindo L. and Li, Lei}, booktitle = {The 43rd International Conference on Machine Learning (ICML)}, location = {Seoul, South Korea}, year = {2026}, month = jul, }