research

current focus

• cybersecurity and AI, and the places where the two meet
• mechanistic interpretability, working out what a model actually computes internally
• compilers

BEACON

Behavioral Engine for Authentication & Continuous Monitoring.

BEACON asks whether the way a person uses their computer can identify them reliably enough to authenticate them continuously, for as long as they are at the machine rather than once at a login screen. We studied this through mouse and keyboard behaviour under real cognitive load.

No existing dataset captured that behaviour at the scale or fidelity we needed, so we built one. Over six months we ran a collection lab (Lab L301) at Thapar, where participants played competitive VALORANT while we recorded mouse dynamics, keyboard events, network traces, gameplay video, hardware metadata and in-game configuration, all synchronised. VALORANT was chosen because it is demanding, multiplayer and free of sensitive personal data.

• 445 GBreleased data
• 102.5 hrsactive capture
• 79sessions
• 28participants
• 90.2Mmouse events
• 57.1Mnetwork packets

Figures from the dataset's release metadata on Hugging Face.

The paper, the dataset and the logger are all public. We welcome collaboration on behavioural biometrics, continuous authentication and anti-cheat research.

my part

• Provided the GPU machines the VALORANT sessions ran on.
• Supervised the collection lab whenever I was available.
• Built the data pipeline and the backend server participants used to upload their session data.
• Aggregated, cleaned and released the dataset on Hugging Face.
• Co-authored the paper.

• BEACON: A Multimodal Dataset for Learning Behavioral Fingerprints from Gameplay Data Paper

  arXiv preprint · 2026
  Ishpuneet Singh, Gursmeep Kaur, Uday Pratap Singh Atwal, Guramrit Singh, Gurjot Singh, Maninder Singh

  arXiv:2605.10867 · Hugging Face

• BEACON Dataset Dataset

  Hugging Face · 2026
  445 GB of anonymised multimodal behavioural data under CC BY-NC 4.0. Aggregated, cleaned and released by me.

  Hugging Face

• BEACON Logger Software

  Zenodo · 2026
  The instrumentation used to capture every modality in sync over six months.

  Zenodo

Mentored by Dr. Maninder Singh (CSED, TIET) and Gurjot Singh. Special thanks to Akshat Kaushik, Ishika Walia, and Rudraksh Chugh for monitoring the lab for most of the six-month campaign.

honors

• Grand Finalist, Bharatiya Antariksh Hackathon 2025 2025

  Indian Space Research Organisation (ISRO)

  Team Amrit was one of the top 30 teams out of more than 8,000 and presented at the grand finale at NRSC Hyderabad. We worked on identifying halo coronal mass ejection events from SWIS-ASPEX particle data aboard Aditya-L1. We used concept bottleneck models so that each forecast explains its reasoning instead of only giving an answer.

• Runner-up (2nd place), Synthetic Data AI Agents & OpenEnv Challenge 2025 2025

  AMD · Unsloth · PyTorch

  Five days of fine-tuning models into questioner and answer agents, which then competed against other teams' agents in a bracket tournament.

• First runner-up, Smart Punjab Hackathon 2022

  Thapar Institute of Engineering & Technology

  Team SUNS, for Smart Switch, out of more than 300 participants.

profiles

ORCID 0009-0009-7432-356X · Google Scholar · Hugging Face