Michael Puodziukas
AI Evaluation Engineer
Remote (US), based in Arizona - michael@puodziukas.dev - puodziukas.dev -
github.com/mpuodziukas-labs
Mid-level and senior IC roles. W2 full-time or contract-to-hire, Remote (US), based in Arizona.
W2 full-time or contract-to-hire. Remote (US), based in Arizona. Available now; start date set at
offer. Domains: agentic AI, AI security / red-teaming, inference infrastructure, mainframe
modernization.

SUMMARY
Independent engineer since 2023, building AI evaluation and security tooling; public repositories
published October 2026. Tests that catch AI model failures before release, published so anyone can
run them.
The LLM adversarial gate below caught 171/177 adversarial attempts and missed 6/177, with 4/107
benign prompts over-blocked, on its own synthetic test corpus (n=284, written for the repo). Every
result ships as clone-and-run code.

EXPERIENCE
AI Evaluation Engineer | Self-employed (Independent) | 2023 - Present
Build AI systems and red-team them as the same person; every result ships as clone-and-run code.
- Agentic AI systems: design, build, and adversarial validation. Built a release gate that replays a
  fixed synthetic corpus of adversarial and benign prompts before each release, so a regression is
  caught before it ships; the repo is public and CI fails if the corpus differs from a fresh build.
- AI security: LLM red-teaming and guardrails. Authored an open-source adversarial gate
  (llm-adversarial-gate): a deterministic, rule-based check, not a model call. On its own synthetic
  test corpus (n=284, written for the repo): 171/177 adversarial attempts caught, 6/177 missed
  (published known gaps), 4/107 benign prompts over-blocked. Clone and run. Designed
  defense-in-depth for indirect prompt injection and agent hijacking; each defense paired with an
  adversarial replay case so coverage is demonstrable, not asserted.
- Mainframe forensics: COBOL arithmetic integrity. Published an open-source COBOL arithmetic
  forensic tool (cobol-pic-probe): a fixed-point truncation probe. Replay on 47,312 synthetic
  records reproduces a $40,832.41 synthetic compounding loss. Clone and run.
- Infrastructure: self-hosted inference (home lab). Built and operate a 2-node self-hosted inference
  cluster at home. No public replay. Walkthrough on request.

PROJECTS
Verify in 60 seconds: clone a repo below and run the command beside it.
- https://github.com/mpuodziukas-labs/cobol-pic-probe (published 2026-10-03): COBOL forensics:
  fixed-point truncation probe. Run: python3 probe.py
- https://github.com/mpuodziukas-labs/llm-adversarial-gate (published 2026-10-03): AI security and
  evals: adversarial release gate. Run: python3 build_corpus.py && python3 evaluate.py
- https://github.com/mpuodziukas-labs/rag-groundedness-gate (published 2026-10-02): RAG evaluation
  and groundedness guardrails.
- https://github.com/mpuodziukas-labs/agent-trace-receipts (published 2026-10-02): Agent
  observability and audit trail.
- https://github.com/mpuodziukas-labs/attested-tool-results (published 2026-10-06): Agent security:
  signed, fresh, single-use tool results.
- https://github.com/mpuodziukas-labs/taint-flow-guard (published 2026-10-06): Prompt-injection
  defense: taint flow.
- llm-adversarial-gate result: 171/177 caught, 6/177 missed (synthetic corpus). Run: python3
  build_corpus.py && python3 evaluate.py

TECHNICAL SKILLS
Languages: Python - COBOL (read and probed via cobol-pic-probe) - C (no public artifact yet)
Evaluation and guardrails: RAG evaluation - groundedness guardrails - adversarial release gate
Agent security and observability: agent observability - prompt-injection defense - OWASP LLM Top 10
(five classes)
Mainframe / Legacy: COBOL forensics with a Python and pytest replay probe (cobol-pic-probe)

EDUCATION
University of South Florida, B.S. Entrepreneurship
Western Governors University, Computer Science coursework (no degree)
Relevant coursework (incl. transfer credit): Calculus I-II, Discrete Math I, Trigonometry,
Statistics I-II, Data Structures and Algorithms I, Version Control, intro Python and Java, Web
Development, Networking, Relational Databases, Database Programming and Management. Self-study:
Discrete Math II, Linear Algebra, DSA II, software QA, Harvard CS50 (no certificate), Helsinki Java
MOOC.