Skip to content
View AUSTIN-OMONDI's full-sized avatar
😊
I may be slow to respond.
😊
I may be slow to respond.

Block or report AUSTIN-OMONDI

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.

Content in all repositories owned by your account will be closed.
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
AUSTIN-OMONDI/README.md

Austin Omondi

AI red teamer working on LLM and agentic systems. I build these systems, then break them — the engineering side is where the findings come from.

Nairobi, Kenya (UTC+3) · Blog · LinkedIn · omondiaustine2@gmail.com


Worth clicking

vuln-langgraph-001 — A deliberately vulnerable three-agent LangGraph system. Ten documented vulnerabilities, each with a working PoC, a framework mapping, and a remediation: cross-agent injection, confused-deputy tool-argument injection, goal hijacking, state pollution, path traversal, system-prompt leakage.

vuln-rag-001 — An intentionally vulnerable RAG service. A poisoned document planted through an unauthenticated endpoint leaks a system-prompt secret from an entirely benign user query. Attacks are automated and scored with PyRIT and Garak, including a false-positive analysis of the automated scanner.

GenAI-Agent-Security-Initiative — My LangGraph code sample for ASI02 (Tool Misuse & Exploitation), submitted to the OWASP Top 10 for Agentic Applications.


Currently

Working through a structured 26-week AI security program in the open at ai-sec-journey — adversarial ML, agentic security, and paired secure/insecure builds.

Competitive red teaming: Gray Swan Arena, Lakera Gandalf (top 8%), Hacker101 (36 points — web and crypto, including a CBC padding oracle written from scratch).

The older repositories here are data science and client engineering work from 2023–2025, before I moved to security full time.


Open to

Subcontract engagements with security consultancies and independent consultants:

  • adversarial testing of LLM, RAG and agentic systems
  • secure code review of AI application code and retrieval pipelines
  • turning raw findings into client-ready reports

White-label, and small fixed-price pilots are welcome. A sample report is available on request.

Full-time roles in AI red teaming, LLM security research and agentic security engineering.

Remote, UTC+3 — full overlap with European teams, afternoons with US Eastern. omondiaustine2@gmail.com

Pinned Loading

  1. vuln-rag-001-ai-redteam-lab vuln-rag-001-ai-redteam-lab Public

    Vulnerable RAG + tool-enabled agent (FastAPI, LangChain, ChromaDB, gpt-4o-mini). Poisoned doc was retrieved into context but refused by the model: refusal is not a control. Unrestricted read_file t…

    Python

  2. vuln-langgraph-001 vuln-langgraph-001 Public

    Deliberately vulnerable 3-agent LangGraph system: 10 documented vulnerabilities with working PoCs, framework mappings and remediations.

    Python

  3. GenAI-Agent-Security-Initiative GenAI-Agent-Security-Initiative Public

    Forked from GenAI-Security-Project/GenAI-Agent-Security-Initiative

    My fork of the OWASP agentic security code-sample project - LangGraph ASI02 (Tool Misuse) sample, submitted upstream as PR #18.

    Python 1

  4. health-insight-research-agent health-insight-research-agent Public

    Multi-agent AI system that generates evidence-based medical reports for Kenyan healthcare settings using LangGraph, OpenAI, and Tavily.

    Python 1

  5. ai-sec-journey ai-sec-journey Public

    Structured 26-week AI security program in the open: adversarial ML, agentic security, paired secure/insecure builds.

    Python

  6. Indirect-Prompt-Injection-via-Retrieved-Content Indirect-Prompt-Injection-via-Retrieved-Content Public

    How Untrusted Data Can Control an AI Assistant