profile image

Neeladri Bhuiya

About Me

I am a Senior Machine Learning Engineer at A10 Networks, where I work on AI safety and develop methods for detecting vulnerabilities in codebases.

Before this, I was a master’s student at the Manning College of Information and Computer Sciences at UMass Amherst, broadly interested in AI Safety, representation learning, and LLM evaluation. My master’s thesis, supervised by Dr. Haw Shiuan Chang and Dr. Andrew McCallum, proposes modeling LLM prompts as box embeddings, axis-aligned hyperrectangles in high-dimensional space. Unlike vector embeddings, which capture only topical similarity, box embeddings jointly represent both semantic relevance and specificity, better preserving prompt structure and revealing improved performance and interesting scaling behaviors. The paper is available here: Prompt2Box.

During my undergrad, I worked with Dr. Viktor Schlegel and Dr. Stefan Winkler on multi-hop reasoning in LLMs. I designed an adversarial attack using dependency parsing to expose systematic shortcutting behavior during inference, finding that LLMs exploit such shortcuts more covertly than smaller models like BERT and Longformer. This work was published at EMNLP 2024.

News

  • November 2026 — Serving as a reviewer for ICLR 2027.
  • September 2026 — Received a Top Reviewer Award at NeurIPS 2026.
  • August 2026 — Prompt2Box was accepted to the EMNLP 2026 main conference.
  • May 2026 — Joined A10 Networks as a Senior Machine Learning Engineer.
  • January 2026 — PLAGUE was accepted to ICLR 2026.
  • May 2025 — Joined A10 Networks in San Jose as an AI Safety Intern.
  • November 2024 — Presented Seemingly Plausible Distractors at EMNLP 2024 in Miami.
  • September 2024 — Began my M.S. in Computer Science at UMass Amherst.
  • May 2024 — Graduated from NUS with a B.S. in Computer Science.

Selected Publications

View all
EMNLP 2026 (Main)

Prompt2Box

Neeladri Bhuiya, Shib Sankar Dasgupta, Haw-Shiuan Chang, Andrew McCallum

Box embeddings capture both semantic relevance and specificity, exposing more fine-grained patterns in LLM behavior than vector embeddings.