system online · cf 979 (max 1014) · 11 projects · new delhi, in · --:--:-- ist

// hello there, i am

UTSAV GUPTA

▊

I break, build and secure software — from LLM red-teaming frameworks and a BLAKE3-based CSPRNG, to game engines in C and full-stack apps on the edge. B.Tech CSE '28 at VIPS, New Delhi.

// about me

whoami

I'm Utsav Gupta, a computer science undergrad who lives at the intersection of offense and defense. My newest work is SAAGA — an automated red-teaming framework for large language models, built while going deep on LLM security. Before that I engineered a hybrid CSPRNG on BLAKE3 that passes all 17 NIST randomness tests, wrote a 2D game engine in C/SDL2, and interned in cybersecurity & digital forensics under ISEA Phase III (MeitY).

When I'm not auditing randomness or mutating jailbreak prompts, I'm running LLM inference — Qwen3 27B, Llama 70B and flash variants through vLLM, llama.cpp and Ollama — learning CUDA programming and deep learning, or harnessing agents with my own OMP setup.

focusLLM inference · security · systems
currentlyCUDA programming · deep learning
inference stackvLLM · llama.cpp · Ollama
models runqwen3 27b · llama 70b · flash
osCachyOS · i use arch btw
languagesPython · C · C++ · TS · Java · SQL

// software

featured_projects

Eleven builds I'm proud of — security tools, cryptography, engines, and systems. Everything else lives on GitHub.

// competitive programming & open source

the_arena

codeforces — rating historyprofile ↗
–current
–max rating
–contests
Utsav Gupta
@Utsav-X-Bit
Questions Solved–
Active Days–
You can find me on ...
Utsav Gupta
@Utsav-X-Bit
Contributions–
Active Days–
Public Repos–
You can find me on ...
github — contribution activityprofile ↗
– contributions– active days
lessmore
codeforces — submission activityprofile ↗
– submissions– active days
lessmore
hpc — compute used for SAAGAmacchina

// experience & education

the_journey

now

LLM Inference & GPU Programming

Self-directed · current focus

Running local inference on qwen3.8-27b, qwen3.6-27b, llama-70b and qwen3.8-flash through vLLM, llama.cpp and Ollama. Learning CUDA programming and deep learning, and harnessing agents with a self-curated OMP setup.

vLLMllama.cppOllamaCUDAdeep learningagent harnessing
jun 2025 — jul 2025

Cybersecurity & Digital Forensics Intern

ISEA Phase III · MeitY · IGDTUW, New Delhi

Investigated ~700 GB of forensic images from 26 seized HDDs in the NEET 2017 case. Built a unified SIEM with ML-based anomaly detection (C++ log-parsing core, Python analytics) and completed CHFI-aligned digital forensics training.

digital forensicsSIEMC++ML
2024 — 2028

B.Tech, Computer Science & Engineering

Vivekananda Institute of Professional Studies, New Delhi — CGPA 8.95/10

Coursework across data structures, DBMS, discrete math and ML foundations, alongside self-directed security work and competitive programming.

DSADBMSOSML
2022 — 2024

Class XII · Class X

Banasthali Public School, New Delhi — 88% (XII) · 91% (X)

// credentials

certifications

Verified certificates & achievements — click any card to view the original document.

// interactive

try_the_shell

A working shell for this portfolio — browse projects, switch the phosphor, dig up secrets. Start with help.

utsav@x — portfolio shelltype help

// contact

init_handshake

Open to internships, security projects, and interesting builds. My inbox is the fastest route — I reply within a day.