0ximu | Security Researcher

⠀⠀⠀⠀⣠⣤⣤⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀▌⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀◙█⠀⠀▀⢀⣤⣤⣄⠀⠀⠀⠀ ⠀⠀⠀⠸⣿⣿⣿⣿⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣿⣿⣿⣿⣇⠀⠀⠀ ⢀⣤⣤⣼⣿⣿⣿⣿⠃⠀⠀⠀⠀⠀⠀⠀⠀⣀⣤⣶⣾⣿⣿⣿⣿⣿⣿⣷⣶⣤⣄⠀⠀⠀⠀⠀⠀⠀⠀⠀⢿⣿⣿⣿⣿⣤⣤⡀ ⣿⣿⣿⣿⣿⣿⣿⣿⣦⡀⠀⠀⠀⠀⢀⣴⡞⠹⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⡿⣦⡀⠀⠀⠀⠀⢀⣴⣶⣿⣿⣿⣿⣿⣿⣿ ⠙⠿⣿⡿⠏⢿⣿⣿⣿⣿⣦⣀⠀╝⣿⠟⣤⡿⢁⣿⢛⣿⣿⣿⣿⣿⣿▓⣿⡈⢿⣤⡙⣿⣦⡀⢀⣰⣿⣿⣿⣿⡿⠙⢿⣿⠿⠋ ⠀⠀⠀⠀⠀⠀⠙⠻⣿⣿⣿⠃⠺⠛⠟⠻⠛⠓⠛⠛⠛⠛⠛⠛⠛⠛⠛⠟⠛⠟⠆⠟⠛⠚⠟⠳⠘⣿⣿⣿⡿⠋⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠈⠻⡏⠀⡀⠂⠌⠀⠌⠠⠁⠌⠠⠁⠌⠠⠁⠌⠐⡀⠡⠀⠂⠈⠐⠀⢂⠐⠀⢸⠟⠉⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⢀⣀⣀⣀⣀⣀⣀⣀⣀⣁⣐⣈⣀⣀⣐⣀⣀⣂⣐⣀⣈⣀⣂⣀⣐⣀⣂⣁⣀⣁⣂⣀⣁⣀⣀⣀⣀⣀⣀⣀⠀⠀⠀⠀ ⠀⠀⠀⠀⠻⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠿⠇⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢿⣿⣿⣿⣿⡿⠋⠉⠛⠻⣿⣿⣿⣿⣿⣿⠟⠉⠉⠛⠿⣿⣿⣿⣿⣿⠃⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠘⣿⣿⣿⡟⠀⠀⠀⠀⠀⠀⢻⣿⣿⣿⠁⠀⠀⠀⠀⠀⠹⣿⣿⣿⠏⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠘⣿⣿⡇⠀⠀⠀⠀⠀⠀⣼⣿⣿⣿⠀⠀⠀⠀⠀⠀⢠⣿⣿⠏⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠘⢻⣿⣦⣀⣀⣀⣠⣴⣿⡿⠿⣿⣷⣄⣀⣀⣀⣤⣿⠟⠁⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠉⠻⢿⣿⣿⣿⣿⣇⠀⠀⣨⣿⣿⣿⣿⠿⠋⠁⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⣠⢰⣶⣬⣍⢙⣛⠛⠙⠋⠙⣉⡉⣭⣴⣾⡄⣦⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⣴⣿⡇⣍⡛⠿⠏⣷⣿⣿⣿⢸⣿⣿⣧⢿⠿⠟⣁⢻⣿⣦⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣀⣴⣿⣿⣿⡇⠾⣿⣷⣆⣤⣭⣭⣍⢨⣭⣭⣥⠐⣶⣿⡟⢸⣿⣿⣿⣦⣄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⢠⣴⣾⣿⣦⣼⣿⣿⣿⣿⠟⠃⣷⣬⣙⡛⠻⠿⣿⣿⢸⣿⡿⠿⢘⣋⣥⣶⠘⠻⣿⣿⣿⣿⣷⣴⣾⣿⣶⡄⠀⠀⠀⠀ ⠀⠀⠀⠀⢺⣿⣿⣿⣿⣿⣿⣿⠟⠁⠀⠀⢻⣿⣿⣿⣷⣶⣶⣶⣶⣶⣶⣾⣿⣿⣿⡏⠀⠀⠈⠻⣿⣿⣿⣿⣿⣿⣿⡿⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠙⠛⢻⣿⣿⣿⣿⡆⠀⠀⠀⠀⠻⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⠟⠀⠀⠀⠀⢠⣿⣿⣿⣿⡟⠛⠋⠁⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠸⣿⣿⣿⣿⠇⠀⠀⠀⠀⠀⠈⠿⣿⣿⣿⣿⣿⣿⣿⣿⠿⠁⠀⠀⠀⠀⠀⠈⣿⣿⣿⣿⡇⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠙⠛⠛⠁⠀⠀⠀⠀⠀⠀⠀⠀⠀⠉⠉⠛⠛⠉⠉⠀⠀⠀⠀⠀⠀⠀⠀⠀⠈⠛⠚⠋⠀⠀⠀⠀⠀⠀⠀⠀

██████╗ ██╗ ██╗██╗███╗ ███╗██╗ ██╗ ██╔═══██╗╚██╗██╔╝██║████╗ ████║██║ ██║ ██║ ██║ ╚███╔╝ ██║██╔████╔██║██║ ██║ ██║ ██║ ██╔██╗ ██║██║╚██╔╝██║██║ ██║ ╚██████╔╝██╔╝ ██╗██║██║ ╚═╝ ██║╚██████╔╝ ╚═════╝ ╚═╝ ╚═╝╚═╝╚═╝ ╚═╝ ╚═════╝

> RESEARCH

Security findings, write-ups, and disclosed vulnerabilities

$ cat focus_areas.txt

LLM Security

Prompt injection, jailbreaks, guardrail analysis. How do you make AI do what it shouldn't?

AI-Native Attack Surfaces

Every tool plugged into an LLM is a door. I check if they locked it.

Semantic Analysis

Using understanding/compression tools for vulnerability discovery. If you can map meaning, you can map flaws.

[QUEUE] Pending Disclosures

[PENDING]

[REDACTED]

2025-XX

Responsible disclosure in progress. Good things come to those who wait. And report responsibly.

> PROJECTS

Things I'm building and breaking

MU

Machine Understanding

[RESEARCH][ACTIVE]

Semantic compression for codebases. Ask a question, get exactly the code that matters. 92-98% compression. Because LLMs shouldn't read your entire repo to answer one question.

$ cat mu/architecture.md

- Graph-first design with .mubase storage (DuckDB)
- Triple-layer: graph + vector + temporal
- MUQL query language for surgical context extraction
- Core: get_context_for_question() — ask a question, get exactly the code context needed

$ git status
On branch: divine-vision-v2

[ GitHub → ]

Recon Toolkit

[INTERNAL][PRIVATE]

Custom automation for bug bounty recon. Details classified. Results: not.

$ ./recon.sh --target <redacted>

[*] Subdomain enum: done
[*] Port scan: done
[*] Tech fingerprint: done
[*] Output: ./targets/

$ ls -la ~/projects/

total 2 entries

drwxr-xr-x 2 0ximu 0ximu 4096 Dec 2025 ./

> ABOUT

> CONTACT