Skip to content
A blacksmith's shop at night, a bar of white-hot iron on the anvil the only source of light, throwing hard shadows across the smiths, apprentices, and the timber frame of the shop.

Software & Security Engineer — B.Tech (Cyber Security) @ VIT Chennai

Rakshit Rameshbabu

I build full-stack products and backend systems, taken end-to-end.

verified record
systems built end-to-end — requirements → CI
3
CGPA / 10 — VIT Chennai
9.07
analyzers in Warden, my supply-chain security platform
14
Read the Warden case fileRésumé

act 01 — the record

An alchemist kneeling alone at night in a vaulted room, hands raised before a flask that has begun to glow.

Software engineering, security, and applied ML.

End to end, from written requirements to CI-tested deployments.

B.Tech Computer Science (Cyber Security) student at VIT Chennai, CGPA 9.07. I build full-stack products and backend systems and take them end-to-end — from written requirements and architecture to CI-tested, reproducible deployments.

My work sits at the intersection of software engineering, security, and applied ML: a supply-chain security platform that checks Python packages without running them, a published-grade evaluation study of ML job scheduling, and a cross-platform wellness app. I work AI-native — LLM coding tools are part of my daily loop for prototyping, debugging, test generation, and review.

Interests

  • scalable backend systems
  • applied AI/ML
  • security research
  • post-quantum cryptography

act 02 — the work

A working forge at night, a white-hot ingot on the anvil throwing hard light across the smith, his family, and the timber frame of the shop.

Security checks for Python packages

Judges what a package would actually do.

Gives a Python package an allow, warn or block verdict — from what its code does, where it came from, and what is known about it.

analyzers
14
detected (synthetic)
13 / 14
backend tests in CI
2,908
install-time download and execute (synthetic)→ risk 100 — blocktyposquat of requests (synthetic)→ risk 47 — warnSSH private key read and sent over a socket (synthetic)→ risk 33 — allow (missed)
FROM WARDEN'S BENCHMARK REPORT — 3 OF 22 SYNTHETIC SAMPLES, THE MISS INCLUDED

a synthetic benchmark, not a real-world detection rate: 13 of 14 malicious samples detected, 1 of 8 harmless look-alikes flagged, none of the 8 blocked.

Read the case fileRepository

act 03 — warden

A philosopher lecturing on a brass orrery, a lamp standing in for the sun at its centre and lighting the listening faces from below.

Can the ML scheduler tell same-size jobs apart? No.

The answer is a proven negative result.

An evaluation study of ML-based GPU-cluster job scheduling — and a proven negative result.

counterexamples
0
dispatch instants
45,432
paired TOST
p = 2.6×10−16

Sorting by requested size is statistically equivalent to the full XGBoost pipeline.

the learned score is a function of requested job size alone.

Read the case fileRepository

act 04 — the scheduler

Two girls bent over a kitten by candlelight, absorbed in a small domestic ritual repeated night after night.

One app for plants, fitness, and food

No second implementation to drift.

Cross-platform app unifying plant care, fitness, and nutrition in one daily dashboard with streaks, reminders, and cloud sync.

functional requirements
228
user stories
119
use cases
89

the web app is live on GitHub Pages, the API deploys via a one-click Render Blueprint, and all 307 tests run in CI on every push and PR.

Read the case fileRepository

act 05 — plantpal+

Surgeons crowded around a dissection table as Dr Tulp lifts the tendons of a forearm with forceps, everyone watching the evidence rather than the body.

The ML scheduler doesn't beat a simple size sort.

The honest baseline is not FIFO.

From the Proactive Feasibility Scheduler study — a proven negative result, verified across 45,432 dispatch instants on the reference platform (41,786 real, 3,646 synthetic). LaTeX manuscript in progress.
a wait-time feature set can only produce a meaningful ranking if it contains a per-job attribute that is not a function of size given the [cluster] state … Phase D shows that condition is necessary but not sufficient … report the ML-free control the feature set implies, not FIFO.
Mean wait by policy — SDSC SP2 trace, second-exact replay · simulated minutes, lower is better · source: 05_results/trace_schedulers

Proactive (XGBoost) ties its own ML-free control, Smallest-first — the degeneracy, visualized.

Mean wait by policy — SDSC SP2 trace, second-exact replay — simulated minutes, lower is better
PolicyMean wait
SRPT (oracle, preemptive)60.1
SJF (oracle)110.8
SJF (real user estimates)115.8
Proactive + estimate feature125.0
HRRN (real estimates)126.3
Smallest-first (no ML)145.0
Proactive (XGBoost)145.0
FCFS + first-fit174.6
EASY backfill (oracle)183.7
EASY backfill (real estimates)195.1
Conservative backfill207.5
Strict FIFO755.1
View the study

act 06 — the finding

A river valley under a full moon, the water carrying a cold band of reflected light between dark banks.

Everything public lives on GitHub.

Skills

Languages
  • Python
  • Java
  • C/C++
  • TypeScript
  • JavaScript
  • SQL
  • HTML/CSS
Core CS
  • Data Structures & Algorithms
  • OOP
  • Operating Systems
  • Computer Networks
  • DBMS
Backend & Frameworks
  • FastAPI
  • Node.js/Express
  • REST APIs
  • SQLAlchemy
  • React
  • React Native (Expo)
  • Next.js
  • scikit-learn
Databases & Infra
  • PostgreSQL
  • Redis
  • SQLite
  • Docker
  • Git/GitHub
  • GitHub Actions (CI/CD)
  • AWS (basics)
  • Linux
Security
  • Wireshark
  • Nmap
  • Burp Suite
  • Metasploit
  • Nessus
  • ClamAV
  • network/web security
  • threat detection
  • supply-chain security
AI-Native Workflow
  • LLM-assisted development (Claude Code)
  • prototyping
  • debugging
  • test generation
  • code review

Achievements

  • First Prize — Cyber Secure 360 Expo 2025

    Jun 2025

    Organized by SCOPE, VIT Chennai.

  • Top 100 Teams — FarAway Zuup Hackathon

    Jun 2026

    Among ~11,000 participants.

Education

  • B.Tech CSE (Cyber Security)

    VIT Chennai

    2024–Present

    9.07

    CGPA / 10

  • Class XII (PCM-CS)

    National Public School

    2023–2024

    93.6%

  • Class X

    National Public School

    2021–2022

    96.2%

Certifications

  • Microsoft Certified: Azure Fundamentals

    Rakshit Rameshbabu — reg. D85FB0-05V137

    Microsoft Certified: Azure Fundamentals, issued by Microsoft.

    • 24 July 2026

  • Certificate of Appreciation

    Rakshit Rameshbabu — reg. 24BYB1117

    Awarded for securing First Prize at Cyber Secure 360 Expo – 2025, organized by School of Computer Science and Engineering (SCOPE), Vellore Institute of Technology, Chennai.

    • 21 June 2025
    • Dr. Karmel A — HOD, B.Tech CSE Cyber Security
    • Dr. Ganesan R — Dean, SCOPE

Coursework

  • AI Fluency: Framework & Foundations

    Rakshit R

    Certificate of Completion, issued by Anthropic. Delivered in partnership with University College Cork (Ireland), Ringling College of Art + Design, Higher Education Authority (An tÚdarás um Ard-Oideachas), National Forum for the Enhancement of Teaching and Learning in Higher Education.

  • Claude 101

    Rakshit R

    Certificate of Completion, issued by Anthropic.

  • Claude Code in Action

    Rakshit R — reg. k66h6cj46fuo

    Certificate of Completion, issued by Anthropic.

  • Introduction to Model Context Protocol

    Rakshit R

    Certificate of Completion, issued by Anthropic.

Archive

  • Sep 2026

    Fillwright — Local Resume Autofill

    Chrome extension that reads a resume on-device, builds an editable profile, and fills job application forms after showing exactly what it will write. It never submits an application or guesses work authorization, visa status, demographics or salary. Covered by 96 end-to-end tests in real Chrome. MIT licensed.

    • TypeScript
    • React
    • pdf.js
  • Jul 2026

    Taintwall — AI Agent Tool-Boundary Firewall

    In-process provenance and policy firewall for AI agent tool boundaries, defending against indirect prompt injection; four defense layers (Unicode/markup normalization, content-signal classifier, intent-gated policy, argument-level provenance) plus an attack harness and labelled corpus — measured 43% → 0% exfiltration with benign utility intact.

    • Python
  • Jul 2026

    DocForge — Personal Document Studio

    Client-side document studio that turns Markdown into typeset PDFs and Word documents — a live paginated preview that is itself the editing surface, automatic table of contents with real page numbers, per-page footnotes, numeric or author–year citations, cross-references, and LaTeX math exported to Word as real editable equations — in one self-contained HTML file: no account, no server, works offline. MIT licensed.

    • JavaScript
    • KaTeX
  • Aug 2025

    SentinelCore — OS-Level Malware Analysis Framework

    Kernel-level malware analysis with real-time syscall monitoring (strace/eBPF); container-free isolation via Linux namespaces and cgroups; multi-engine detection pipeline (entropy analysis, ClamAV, VirusTotal API, AI-assisted log analysis).

    • Python
    • C
    • Linux
    • eBPF
    • 2025-08
    • completed
    • python · c · eBPF
  • Dec 2025

    Web Application Security Suite

    Security testing and monitoring platform: OWASP Top 10 vulnerability scanner (SQLi, XSS, CSRF), request monitoring with anomaly detection and rate limiting, JWT-based Zero Trust auth, attack simulation for testing.

    • Python
    • JavaScript
    • 2025-12
    • completed
    • python · javascript

  • Full archive

    More repositories — everything public lives on GitHub.

    Browse all repositories

act 07 — the ledger

A woman shields a single candle flame with her cupped hand as a young girl reads beside her, both faces lit by the flame alone.

Open to internships and full-time roles in software and security engineering.

Based in Chennai, India. The fastest way to reach me is email — copy it here, or find me on GitHub and LinkedIn.

rakshitoffl@gmail.com
Download résuméGitHubLinkedIn

Thank you for reading this far.

act 08 — the close