Vaghesan Sundaram

Cybersecurity intern and Computer Science and Mathematics student at the University of Maryland building Python systems across application security, least-privilege authorization, cyber threat intelligence, authentication, and robot-policy evaluation.

CompTIA Security+ certified. U.S. citizen. Expected graduation: December 2027. Based in College Park, Maryland.

Experience

Incoming Technology Intern (TIP)

Capital One

Cybersecurity Intern

Redmind Technologies

  • Identified and documented 7 reproducible findings across authentication, data exposure, configuration, and dependency risk by combining Semgrep SAST with manual secure code review of a public-facing client application.
  • Delivered a developer-facing report and slide briefing that validated evidence, prioritized all 7 findings by severity, and mapped each issue to actionable remediation guidance for developer implementation and client review.
  • Developing a PyTorch pipeline with MuJoCo, robosuite, robomimic, and LeRobot to measure training-data and reward-label poisoning across behavior cloning, transformer, VLA, and offline RL policies.

Information Technology Intern

New Jersey Courts

  • Installed courtroom recording hardware to capture legal proceedings and support court recordkeeping.
  • Improved IT asset inventory accuracy across 5 departments (Finance, IT, Civil, Probation, and Operations) by cataloging devices and reconciling physical equipment with tracking records for asset control.

Projects

Built a local Python MCP proxy that enforces launch-time filesystem policies for allowed paths, exact arguments, and call quotas. It serializes quota reservations, records redacted audit events, fails closed on errors, and tracks unknown outcomes. In simulation, it blocked 12/12 unauthorized proposals while completing 10/12 attack tasks and 24/24 controls.

Designed a 64-question point-in-time CTI benchmark across 24 dependency groups, with cutoff-aware evidence, authority rules, human review, and source hashes. Constrained output raised evidence binding from 83.3% to 86.5% but lowered exact accuracy from 33.3% to 26.0%; a 144-response follow-up found no material schema benefit.

Designed and built a two-service authentication system that profiles typing cadence and verifies sign-ins, with SQLite persistence, JWT-bound service exchange, rate limiting, fallback authentication, and end-to-end validation.

Designed and implemented Prophet's Python event-to-market pipeline, transforming posts from X into clustered event signals, Grok-generated market questions and probability estimates, and WebSocket actions consumed by downstream exchange components.

Technical Skills

Languages
Python, Java, C++, C, Assembly, Rust
Security
OSINT, Maltego, Shodan, Recon-ng, Maigret, Burp Suite, Nmap, Wireshark, Ghidra, vulnerability triage, TCP/IP, DNS, DHCP, PCAP analysis, Kali Linux, Hashcat, John the Ripper, privilege escalation, exploit development
Tools
Git, GitHub Actions, Docker, Linux, PowerShell, Flask, Pydantic, pytest, Hypothesis, mypy, REST APIs, GDB
Data + ML
PyTorch, MuJoCo, robosuite, robomimic, LeRobot, SQL, SQLite, NumPy, SciPy, scikit-learn