What makes this different from staff augmentation?

One person with seven specializations unavailable in staff aug. COBOL PICTURE-clause forensic find in 127 minutes that 3 years of external audits missed. 3-5x lower cost for broader coverage.

Is the sovereign inference cluster operational?

Yes. 55 commits ahead of upstream on a production fork of exo (upstream lineage). RDMA ring topology. 389ms P50. AWS Bedrock and Azure OpenAI format-compatible. $26,400/mo cloud inference eliminated.

What audit-trail artifacts does the sovereign inference build produce?

SHA-256 hash-chained chain-of-custody on every inference output. Quality-management-system documentation, bias logs, and human-oversight records with a deterministic event-sourced audit trail, operational now.

What compensation are you targeting?

Negotiated privately, by role and scope — remote · W2 · skills-first.

100% remote. US-based, Mountain Time. Async-first. No relocation.

What kind of engineer are you?

Agentic architect and AI/data engineer — agentic systems, sovereign inference infrastructure, and adversarial-AI validation, every result replay-verifiable.

What proof do you offer?

Deterministic replay artifacts. SHA-256 hash-chained chain-of-custody. COBOL PICTURE-clause forensic audit in 127 minutes. 27× throughput. 96% latency cut. 0 false negatives across the public adversarial corpus (140/140 caught, n=220). All verifiable at puodziukas.dev/proof.

Why hire you instead of engaging a consulting firm?

Single operator, full-time, inside your perimeter. The person who scoped the work does the work — every output logged and replay-verifiable.

Yes. Standard MNDA on day one. All work product stays air-gapped.

What stacks do you work in?

COBOL · Rust · Python · TypeScript · Next.js · LLM APIs · AWS · GCP · Terraform · Pulumi · OpenTelemetry · NIST FIPS 203/204/205.

Seeking a full-time W2 role

Remote · U.S.-authorized, no sponsorship · Available immediately

Target roles: AI Security Engineer · Inference/ML Platform Engineer · Verification & Validation · AI Governance · Agentic Systems Architect

47,312 REPLAYS · 0 DIVERGENCE · 55 COMMITS AHEAD OF UPSTREAM · COBOL FIND IN 127 MIN

I build the system and red-team it in the same pass.

The failure class surfaces in the build, not in production. Written scope, first artifact in 48 hours — replay-verified, checkable by your team on its own clock. Senior · remote.

Work with me

Before → After. Every public number has a replay command.

The core claims clone and run from public repos. The cluster reproduces live in a technical screen; NDA evals reproduce on request.

$26,400/mo API →$0/yr API2-node Apple Silicon mesh · zero egress · 389ms P50 · same models, same latency

27% hallucination →<1%adversarial council · 3-lens vote · internal eval n=4,200 — public anchor: 140/140 caught, 0/80 benign over-blocked (n=220)

single-layer baseline →5 / 5 OWASP classesmulti-agent gate vs single-layer on identical test set · zero regressions

5,200 COBOL lines →2,000 Rustbyte-exact parity · 27× throughput · 464 tests · 0 known CVEs · replay-verifiable

I build AI systems and independently prove they work.

One discipline, two tracks: I ship AI systems (Build), and I independently verify & secure them (Assure) — senior-level for either.

BUILD

AI security
inference / ML platform
agentic systems

ASSURE

verification & validation
AI governance
risk

Role fit: the proof, per role

Five target tracks. Every claim is reproducible.

Role	Proof	What it means
AI Security Engineer	COBOL failure class caught in 127 min; public probe reproduces $40,812.81 · 140/140 adversarial caught · 0 CVEs · 5 OWASP-LLM classes	Finds the failure class before production - hands you a clone-and-run probe, not a slide.
Inference / ML Platform Engineer	158 TPS · 389ms P50 · 99.97% uptime · $26,400/mo → $0/yr API on a 2-node Apple Silicon ring · 55 commits ahead of upstream exo	Model-serving that is measured, not estimated - deletes the inference bill with proof.
Verification & Validation	46-gate deterministic battery, each mutation-proven on a planted failure · caught 2 AI false-greens · checksum twin 541,774 files / 0 divergence	Builds the gates that stop bad output shipping - mutation-proven, not asserted.
AI Governance / Risk	audit-trail-by-receipts · provenance · fail-closed leak scanning · EU AI Act + OWASP-LLM mapping	Turns AI behavior into traceable evidence: findings, controls, remediation, residual risk.
Agentic Systems	47,312 deterministic replays · multi-agent council · guardrails · 27% → <1% hallucination (internal eval n=4,200, NDA; public proxy: 140/140 caught)	Evaluates and contains agent failure - replayable, not anecdotal.

AI Security Engineer

Proof: COBOL failure class caught in 127 min; public probe reproduces $40,812.81 · 140/140 adversarial caught · 0 CVEs · 5 OWASP-LLM classes

Signal: Finds the failure class before production - hands you a clone-and-run probe, not a slide.

Inference / ML Platform Engineer

Proof: 158 TPS · 389ms P50 · 99.97% uptime · $26,400/mo → $0/yr API on a 2-node Apple Silicon ring · 55 commits ahead of upstream exo

Signal: Model-serving that is measured, not estimated - deletes the inference bill with proof.

Verification & Validation

Proof: 46-gate deterministic battery, each mutation-proven on a planted failure · caught 2 AI false-greens · checksum twin 541,774 files / 0 divergence

Signal: Builds the gates that stop bad output shipping - mutation-proven, not asserted.

AI Governance / Risk

Proof: audit-trail-by-receipts · provenance · fail-closed leak scanning · EU AI Act + OWASP-LLM mapping

Signal: Turns AI behavior into traceable evidence: findings, controls, remediation, residual risk.

Agentic Systems

Proof: 47,312 deterministic replays · multi-agent council · guardrails · 27% → <1% hallucination (internal eval n=4,200, NDA; public proxy: 140/140 caught)

Signal: Evaluates and contains agent failure - replayable, not anecdotal.

For analyst / GRC / V&V-analyst roles: I produce the reproducible evidence behind every engineering claim - the same work an assurance analyst must do, already done and public.

IN PLAIN TERMS

agentic: AI that runs multi-step tasks on its own
sovereign inference: private, self-hosted AI the organization controls
V&V: testing and proof that the system works
red-team: attack-testing to find weaknesses
deterministic gate: an automated pass/fail check that's identical every run

How it works

01I work from a written spec — no discovery phase, first artifact fast.
02First artifact lands in 48 hours, with its replay command.
03Async-first — code lands in your repo; decisions arrive as written diffs with replay commands your team can verify on its own clock. Low-ceremony, not low-communication.
04Every output is reproducible — clone it and check.

FIRST 30 DAYS

Week 1

First artifact shipped — a deterministic replay command on every LLM decision in the audit trail.

Week 3

Every COBOL arithmetic boundary re-derived. ON SIZE ERROR gaps identified and documented with fix spec.

Day 45

Board question answered with a log line, not a policy document — the replay command in the artifact itself.

Worth a conversation if

+The team runs an agent fleet, an inference bill, or a legacy system that has to be correct under adversarial conditions — and needs one remote IC who ships replay-verified artifacts, not status updates.
+The team measures output by artifacts and replay commands, not standups.
+The role is remote and the scope is written down before day one.

Not a fit if

–The role needs someone to convince the org the problem is real.
–The role is management, not building.
–On-site, daily-standup, ticket-velocity shops.

“What does an internal team not already have?”

structural distance · public proxy: 140/140 caught + $40,812.81 · production record NDA

Structural distance. The probe arrives with no prior attachment to the codebase, so it audits from outside — the team that built a system has a harder time surfacing its own blind spots than someone who didn't. That distance is why a truncation defect can age through years of internal audits that were never designed to surface its class.

“How do you verify the work without a client list?”

replay ships with every deliverable · clone-verify from input hash

The proof is the artifact. Every deliverable ships with the replay command that reconstructs its result from the input hash — clone it, run it, check it against the documented number. The methodology was validated in production before this became a client-facing offering, so what you're auditing is the log, not a reference call.

Open to one Senior role, available immediately. Three questions: the role, the system, the timeline.

GET IN TOUCH · 24-HOUR REPLY

One paragraph. 24-hour reply.

The more specific the note, the more useful the reply.

or email directly →