Manish Kumar Jha
Offensive security engineer and founder of Vyrox Security, currently specializing in AI/LLM security and LLM red teaming. I break into systems (web applications, Active Directory, network attack paths) to understand them, then build my own offensive tooling in Rust and Python. Strong bias toward building over buying, and toward depth over breadth.
Experience
Security Engineer
Vyrox Security
Remote, India
•Open-core autonomous security platform (Rust, Python): HMAC-verified EDR ingestion (CrowdStrike, SentinelOne), deterministic triage, async workers.
•Containment rollback via state machine (executed, rolling-back, rolled-back, failed); pages a human on failure, verified against a mock EDR.
•Per-tenant encrypted EDR credentials, HMAC-signed proxy execution, SHA-256 hash-chained audit log producing a tamper-evident evidence pack.
Security Engineer
Brand 247
Remote
•Authorized web app pentesting on a production ad platform: SQLi, XSS, IDOR, SSRF.
•Built custom tooling to automate testing; reported findings with repro and fixes, and verified them.
Fellow
Superteam India
Remote
- Selected to build in India's Solana ecosystem; shipped a time-locked messaging dApp (School of Solana capstone).
Education
Silicon Institute of Technology (SIT), Bhubaneswar
B.Tech
Bhubaneswar
School of Solana
capstone: Time Capsule Protocol (on-chain time-locked messaging dApp)
Remote
Skills
CERTIFICATIONS & LEARNING
Offensive-security to AI/LLM-security curriculum
Completing a self-directed 26-week offensive-security to AI/LLM-security curriculum.
AI/LLM security specialization
Specializing in AI/LLM security: prompt injection, agentic and RAG attacks, adversarial machine learning, and LLM red-team tooling (garak, PyRIT, promptfoo), mapped to the OWASP LLM Top 10 and MITRE ATLAS, building an AI red-team agent as the capstone.
Offensive Security Studies
Studying (syllabus-based, exams not yet taken): Offensive Security Certified Professional (OSCP, PEN 200), Certified Red Team Professional (CRTP), OffSec OSAI (AI-300).
Projects
Jailbreak-fable
(Javascript)
High-fidelity LLM environment emulation and automated multi-agent jailbreak (Pack Hunt) research lab for adversarial LLM testing.
vyrox-proxy
(Rust)
Hardened containment proxy for approved EDR actions, with HMAC verification, rate limiting, and append-only audit logging.
termino
(Rust)
Decentralized, peer-to-peer communication app using WebRTC, libp2p, and IPFS for encrypted messaging, file sharing, and audio.
nyx
(Rust)
Tor identity rotator with a real-time TUI for per-request circuit rotation and live hop visibility.
deepsentry
(Python)
unsupervised anomaly detection on distributed system logs using deep learning.
Research
(markdown)
Security research lab with box writeups, lab notes, CVE analysis, and blog drafts.
Open-Source
Added Metadata::get_or
dora-rs
Added Metadata::get_or for cleaner parameter access with defaults, reducing boilerplate and improving ergonomics in dora-rs.
Rust Clippy - useless_conversion
rust-lang
Enhanced .into_iter() suggestions for nested references, fixing over-borrowed code patterns in the official Rust linter.