Skip to content
View rvong65's full-sized avatar

Block or report rvong65

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
rvong65/README.md
LinkedIn Banner

I build AI systems people can trust — in security operations and in ethical decision-making.

Trustworthy AI · Responsible / Ethical AI · Cybersecurity · Agentic Systems

I design and ship explainable, safety-constrained AI across two fronts that matter to me equally:

  1. Cybersecurity — systems that detect threats, ground answers in evidence, and refuse to overclaim when the signal is weak
  2. Ethical & responsible AI — tools that audit LLM judgments, stress-test moral consistency, and review agent actions before they execute

Not demos that look smart until you ask why. Every project below is live, tested, and built with production-minded constraints (guardrails, CI, reproducible artifacts).

🚀 What I've shipped

🔐 Cybersecurity & threat intelligence

Project What it does Proof Links
Breach Precursor Detector Anomaly detection over multi-source EDR telemetry (CrowdStrike Falcon, Sysmon, Windows 4688) with confidence gating and SHAP explanations 12 behavioral features · Isolation Forest · 92 CI tests · Dockerized Code · Demo
MITRE ATT&CK Chain Visualizer Groups telemetry into multi-stage attack chains (Execution → Credential Access → Persistence) with explainable confidence scoring 12K+ events → 11K+ chains · live on Streamlit Cloud Code · Demo
Threat Intelligence Assistant RAG over MITRE ATT&CK + CISA KEV with mandatory citations, confidence scoring, and hard abstention 3,312 indexed chunks · 49 tests · 11/12 inspector cases PASS Code · Demo
Simple Autonomous Security Agent Transparent ReAct SOC agent with tool guardrails, private-IP blocking, and a deterministic risk floor 4 read-only tools · 64 offline tests · live demo Code · Demo

⚖️ Ethical & responsible AI

Project What it does Proof Links
Ethical Decision Simulator Stress-tests LLMs on moral dilemmas with framework detection and consistency checks under framing shifts Structured verdicts · crisis-content guardrails · cross-model agreement metrics Code · Demo
Ethical Agent Auditor Real-time review of proposed agent actions via hybrid rules + Chroma RAG + LLM scoring 5 ethical principles · 36 CI tests · Docker + Streamlit Cloud Code · Demo

🛠️ How I build

  • Evidence over vibes — citations required, abstention when retrieval is weak, SHAP and rule-based explanations for analysts
  • Safety by design — input/tool guardrails, rate limits, deterministic floors so models cannot under-rate hard evidence
  • Ethics under scrutiny — framework-aware auditing, consistency checks across rephrased scenarios, structured reports instead of black-box “right answers”
  • Ship it — Streamlit Cloud demos, Docker/Compose, GitHub Actions CI, committed indexes and demo data for clone-and-run

🧰 Stack

Python scikit-learn Pandas Hugging Face Streamlit Docker GitHub Actions

Python · Pandas / PyArrow · scikit-learn · SHAP · FAISS / Chroma · Groq / Ollama · Hugging Face · Streamlit · Plotly · Pydantic · Docker · GitHub Actions

🔭 Currently exploring

Trustworthy & agentic AI · ethical / responsible AI tooling · breach precursor detection · explainable behavioral analytics · LLM consistency and agent-action auditing

📬 Let's connect

If you're hiring for — or building — trustworthy AI for cybersecurity or ethical / responsible AI systems, open an issue on any of the repos above or reach out. I care about systems that hold up under scrutiny.

Pinned Loading

  1. ethical-decision-simulator ethical-decision-simulator Public

    Ethical Decision Simulator: audit how LLMs reason through moral dilemmas—multi-model compare, ethical lens stress tests, semantic agreement metrics, and exportable JSON verdicts. Python, Streamlit,…

    Python 1

  2. ethical-agent-auditor ethical-agent-auditor Public

    Ethical Agent Auditor: review AI agent decisions in real time—hybrid rule checks, Chroma RAG, five-principle LLM scoring, EDS dilemma pairing, and exportable audit reports. Python, Streamlit, Groq …

    Python 1

  3. breach-precursor-detector breach-precursor-detector Public

    Unsupervised anomaly detector that flags early breach precursors (credential dumping, process injection) using Isolation Forest on EDR-style process features. Inspired by CrowdStrike-style EDR — in…

    Python 1

  4. mitre-attack-chain-visualizer mitre-attack-chain-visualizer Public

    Interactive MITRE ATT&CK attack chain visualizer built from real Atomic Red Team telemetry. Groups Sysmon/EDR events into scored process chains with confidence gating and explainable timelines—depl…

    Python 1

  5. simple-autonomous-security-agent simple-autonomous-security-agent Public

    Simple Autonomous Security Agent (SASA): transparent ReAct SOC triage with four read-only tools, guardrails, and risk floor. Python, Streamlit, Groq or Ollama—deployed on Streamlit Cloud.

    Python 1

  6. threat-intelligence-assistant threat-intelligence-assistant Public

    Threat intelligence RAG app: grounded Q&A over MITRE ATT&CK and CISA KEV with mandatory citations, confidence scoring, and query guardrails. LangChain, FAISS, Groq —deployed on Streamlit Cloud.

    Python 1