Proma Roy

Proma Roy

Cybersecurity researcher and adjunct lecturer

I teach computer science and applied mathematics at John Jay College and The City College of New York. My research interests include secure systems, privacy-preserving technology, and equitable computing for underrepresented-language communities.

Email LinkedIn GitHub Google Scholar ORCID
Portrait of Proma Roy

About

Teaching, research, and secure systems.

I am an adjunct lecturer in New York City, teaching introductory programming at John Jay College of Criminal Justice and Applied Linear Algebra at The City College of New York. In the classroom, I use Python-based labs, hands-on exercises, and structured problem solving to help students connect concepts with practical applications.

My recent research examines two areas: privacy-preserving technology for people at risk of intimate partner violence, and the infrastructure barriers that affect AI access for speakers of underserved languages. My technical work also includes applied cryptography, operating-system telemetry, network security, and secure software design.

Before moving into teaching and research, I worked in healthcare credentialing and data operations, where accuracy, documentation, and traceability were central to the job.

Current appointments
John Jay College, CUNY
The City College of New York
Courses, Fall 2026
Introduction to Python
Introduction to Computing & Programming in C++
Applied Linear Algebra
Research areas
Secure systems
Privacy-preserving technology
Equitable and multilingual AI

Experience

Experience.

Current academic appointments and prior professional work.

Current Aug 2026—Present

Adjunct Lecturer

John Jay College of Criminal Justice, CUNY · Department of Mathematics & Computer Science

New York, NY · On-site

Teach undergraduate programming through hands-on activities and projects that make abstract concepts concrete. Courses cover Python data and file processing as well as C++ control structures, functions, arrays, pointers, and structured program design.

  • CSCI 171 · Introduction to Python
  • CSCI 271 · Introduction to Computing & Programming in C++
  • Student-centered learning
Current Aug 2026—Present

Adjunct Lecturer

The City College of New York, CUNY · Department of Mathematics

New York, NY · On-site

Teach Applied Linear Algebra with an emphasis on connecting core mathematical concepts to computational problem solving and practical applications. Support students through Python-based labs, hands-on exercises, and structured exploration of linear algebra concepts.

  • Applied Linear Algebra
  • Python labs
  • Computational problem solving
Mar—Aug 2026

New York, NY · Hybrid

Supported the NSF-funded project Preventing Intimate Partner Violence Using Digital Technology: A Transdisciplinary Approach, contributing to research-driven system design, prototyping, backend data preparation, integration debugging, documentation, and beta-testing readiness.

NSF Partnership for Innovation (PFI-RP) · Multidisciplinary research and nonprofit collaboration

Earlier data and professional experience

Mar 2022—Feb 2025

Associate Vendor Coordinator

Radiant Data Systems, Ltd. · Healthcare credential verification, vendor documentation, and facility-database maintenance with consecutive 100% audit accuracy.

Nov 2021—Mar 2022

Data Artist

Adiva Catalyst Solution & Graphics · Survey logic, data verification, analytical charts, and quality assurance using SPSS and Excel.

Jun 2020—Jun 2021

Junior Data Analyst

a2i, ICT Division, Bangladesh · Data collection, CRM tracking, and documentation for a COVID-19 telehealth and telemedicine initiative.

Selected research

Research publications and presentations.

My recent coauthored work examines the hidden technical costs that underserved-language communities encounter before an AI model ever produces an answer.

arXiv preprintLM4UC @ IJCAI 2026 · Oral + poster

Measuring the Tokenization Premium: A Cost Audit for Underserved Language Communities

Avijit Roy, Proma Roy, Hrishitva Patel · 2026

Introduces the Tokenization Equity Audit (TEA), a reproducible benchmark that evaluates 120 Python-debugging items across Bengali, Hindi, Arabic, Tamil, and Yoruba. The work connects tokenization inefficiency to unequal API cost, effective context capacity, and local deployment overhead — a 128k nominal context window shrinks to roughly 82k English-equivalent tokens for Bengali.

1.56×GPT-4o token premium for Bengali relative to English; Yoruba reaches 2.37×, showing the issue extends beyond script family.
Read on arXiv TEA benchmark Interactive poster LM4UC workshop
arXiv + SSRN preprintILA 2026 · Poster

Structural Silence: When AI Infrastructure Fails Speakers of Underrepresented Languages

Avijit Roy, Proma Roy · 2026

Uses Bengali as a case study to connect four structural barriers: limited web representation, a severe training-token deficit, tokenization overhead, and connectivity exclusion. It argues that dataset scarcity is an infrastructure problem and that offline-first design can be an equity strategy.

<0.5%of global web content is Bengali despite roughly 285 million speakers; the study also documents an illustrative 67:1 English-to-Bengali training-data gap.
Read on arXiv SSRN record Interactive poster Poster DOI

Selected projects

Technical projects.

Selected academic and independent work in secure systems, cryptography, operating-system security, and Bengali language processing.

Detection engineering · 2026

OS Telemetry Evasion & Kernel-Resident Detection

Compared Windows ETW and Linux auditd event streams with independent kernel-level telemetry to identify divergence, ghost processes, and blind spots. The study used Windows callbacks and eBPF tracepoints and validated 12 of 12 Windows tests.

  • Windows 11
  • Ubuntu
  • ETW
  • auditd
  • eBPF
  • MITRE ATT&CK
Applied cryptography · 2025

Threshold ElGamal

A Rust implementation of 3-of-5 threshold encryption with distributed key generation, collaborative decryption, and zero-knowledge proofs.

View repository
Security audit · 2025

Rust E-Voting System Audit

Used STRIDE, OWASP Top 10, CWE mapping, static analysis, runtime fault injection, and build-pipeline inspection to assess an adversarially modified system.

Secure systems · 2025

Smart Thermostat

A secure Rust CLI with Argon2id password hashing, role-based access control, hardened SQLite storage, session controls, audit logs, and integrity checks.

View repository
Network security · 2026

RSA Toolkit + AI Sentinel

A Python secure-channel toolkit for key generation, encryption, signatures, hybrid encryption, controlled attack demonstrations, log triage, and local post-execution analysis.

Research roots · Undergraduate thesis

Bangla Sentiment Analysis

An early NLP research project involving dataset preparation, algorithm development, Python implementation, and sentiment analysis for Bengali text.

Education

Education.

Toolkit & foundations

Skills.

My work combines security methods with clear documentation, reproducible experimentation, and the teaching skills needed to explain why a system behaves the way it does.

Security & systems

Threat modeling · STRIDE · OWASP Top 10 · CWE · secure operating systems · network security · kernel telemetry · applied cryptography · authentication · RBAC

Software & data

Python · C++ · Rust · SQL · SQLite · Git/GitHub · Linux · SPSS · data analytics · integration testing · technical documentation

Credentials & languages

Palo Alto Networks training in cybersecurity, network, endpoint, cloud, and security operations. Bengali (native/bilingual), English (full professional), Hindi (limited working).

Contact

Get in touch.

I welcome conversations about research, teaching, and secure software projects.

[email protected]