Skip to content
View immu4989's full-sized avatar
0 followers · 1 following

Achievements

Achievement: Pull Sharkx2Achievement: Pair Extraordinaire

Achievements

Achievement: Pull Sharkx2Achievement: Pair Extraordinaire

Block or report immu4989

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
immu4989/README.md
imran@vezran: ~/security

LinkedIn X Vezran Zyberpol views

$ whoami

I'm Imran — I build agentic AI for cybersecurity, and I'm allergic to systems that only detect.

At Vezran / Zyberpol I lead AI/ML on a system that takes a threat all the way to closed — with proof. The loop I build by:

detect → decide → approve → act → verify → prove
#  the model reasons · the control layer governs · a human stays in command · every action leaves evidence

Most AI security stops at "we found something." The hard, unglamorous, valuable part is trusted closure — and proof a regulator will accept. That's the part I care about.

I build on the belief that the model is rented and swappable; the control layer is the moat.

$ cat focus.txt

AGENTIC AI      →  multi-agent orchestration, tool use, governed autonomy
LLM SECURITY    →  prompt-injection robustness, red-teaming, adversarial evals
AI SAFETY       →  differential privacy, causal inference, fraud / lure detection
RELIABILITY     →  the eval harness that decides if a model can be trusted
MLOPS           →  notebook → production, without losing the receipts

$ ls arsenal/

stack

+ LangChain · Hugging Face · DSPy · RAG · LoRA/QLoRA · vector DBs

$ ls evidence/  # receipts > résumé — every claim above maps to a repo

Repo Proof of
dspy-security-bench LLM security — does prompt optimization make agents more injectable? Measured against AgentDojo's attack suite.
lurebench AI safety — a benchmark for detecting AI-generated fraud lures (phishing, romance scams).
PyRIT Red-teaming — working in Microsoft's framework for proactively finding risk in generative AI.
opacus · dowhy Rigor — differential privacy for PyTorch, and causal inference done right.

$ ./telemetry --live

stats top langs
streak

$ cat contributions.log  # the threat map

contribution snake
EOF — the model investigates · the control layer decides · a human stays in command · and it proves the threat was closed.

Pinned Loading

  1. dspy-security-bench dspy-security-bench Public

    Measure how DSPy prompt optimization affects the prompt-injection robustness of agentic LLM programs, using AgentDojo's attack suite.

    Python 6

  2. dowhy dowhy Public

    Forked from py-why/dowhy

    DoWhy is a Python library for causal inference that supports explicit modeling and testing of causal assumptions. DoWhy is based on a unified language for causal inference, combining causal graphic…

    Python

  3. FLAML FLAML Public

    Forked from microsoft/FLAML

    A fast library for AutoML and tuning. Join our Discord: https://discord.gg/Cppx2vSPVP.

    Jupyter Notebook

  4. opacus opacus Public

    Forked from meta-pytorch/opacus

    Training PyTorch models with differential privacy

    Jupyter Notebook

  5. PyRIT PyRIT Public

    Forked from microsoft/PyRIT

    The Python Risk Identification Tool for generative AI (PyRIT) is an open source framework built to empower security professionals and engineers to proactively identify risks in generative AI systems.

    Python

  6. lurebench lurebench Public

    A maintained benchmark and evaluation harness for detecting AI-generated fraud lures (phishing, BEC, romance / pig-butchering).

    Python 1