Entropa_

Scott Baker

aimozart · Phoenix, AZ, USA · Remote only · contact through the form below

Forward-Deployed AI Engineer: agentic systems with guardrails, evaluation and signed audit trails · IT, security and regulated data · 12+ years in enterprise IT, security and compliance

GitHub profile: github.com/aimozart

Published on crates.io with Trusted Publishing (GitHub OIDC, no stored tokens): entropa-core · entropa-node · crates.io/users/aimozart

All AI builds and projects here were done with Claude Code on a $20/month Claude Pro plan. This is what is possible from a solo developer with focus and skills.

✓ AWS Certified Solutions Architect – Associate ✓ Databricks Certified Associate Developer for Apache Spark (Scala) ✓ SCCM Administration certificate ✓ crates.io Trusted Publishing Google for Startups Cloud Program

Summary

I'm a forward-deployed AI engineer: I put AI systems into a customer's real constraints and prove what they did. I build agentic systems with the controls around them: rules first, a model second, output guards, a policy gate, a human review step and a signed, verifiable record of every decision. With 12+ years in enterprise IT, security and compliance, I build security-first: audit trails, access control and honest limits are part of the design. My current build is Entropa (entropa.space), a post-quantum-signed, witnessed evidence layer for controls written in Rust, together with a Rust retrieval system over the standards themselves (in progress, not yet professional experience). Every stage has a plan and a forecast, red tests first, an independent review and a measured result, and failed bars are reported as failed: an earlier simulated case study, a claims-triage agent on synthetic data, missed its needs-a-person recall bar on two fresh sets and is kept as a closed case study. I also hold the Databricks Certified Associate Developer for Apache Spark (Scala). AI assistance is disclosed, reviewed and tested.

I work test-first and deterministically: I forecast the outcome, pin every input, confirm a test fails for the stated reason before writing the code, and prove the result against the forecast. The evidence is in my own project, Entropa: a Rust, serverless ledger that signs every AI decision, with over 1,500 automated tests, byte-exact known-answer tests against NIST's official FIPS-204 vectors, fuzzing and mutation testing, and secret scanning in pre-commit and CI with 0 real leaks across the full history. My AI-assisted engineering follows a documented plan, prove, prune cycle in which claims are checked and corrected, not assumed.

I run AI agentically, as a managed team rather than a chat window, and every stage is tested and independently verified before I accept it. The guardrails are code, not promises: a command guard that blocks destructive actions, signed human approvals for anything hard to reverse, and a written record of every decision. That method is how I bring projects from idea to fruition and production: the Entropa ledger (Rust, serverless), the healthcare data platform and the sites, CI, keyless deploys and rollbacks around them. Our latest Paxel report on this AI-assisted work: paxel.ycombinator.com/results/jfx4jvdi.

Before that, 12+ years in regulated enterprise environments: I helped take a healthcare organization to official HITRUST certification (evidence for every control, Qualys and OpenVAS vulnerability scanning turned into remediation plans), ran baseline-configuration audits across a global fleet, wrote the SOPs and remediation documentation that enforced procedure, and administered Windows and Linux systems at scale (Active Directory, Group Policy, SCCM across all its functions; Windows XP → 7 → 10 and 7 → 11 migrations). The same discipline as a regression suite.

I hold the AWS Certified Solutions Architect – Associate and Databricks Certified Associate Developer for Apache Spark (Scala) certifications and an SCCM Administration certificate. Remote only.

Systems administration, the whole way through

Projects

HIPAA-aligned healthcare data pipeline: Scala 2.13 and Spark 4 (synthetic data)

November 2026 · parked as a closed case study · closed case study · repository removed

  • A nightly batch pipeline over synthetic healthcare charts and claims, in four layers (bronze, silver, gold, platinum) on Scala 2.13, Spark 4.1 and Delta Lake: validation with a quarantine invariant (valid plus quarantined always equals input), masking, idempotent reruns and a run record for every night.
  • Safety first: all data is synthetic and handled as if it were real; identifier-shaped values come from reserved invalid ranges, with canary strings and automated leak checks. Controls are mapped to HIPAA, HITRUST-shape and NIST AI RMF (aligned, not certified).
  • A seeded generator of dirty practice data (about 2.9 billion rows at the largest tier, with planted defects and a private answer key), plus a public FastAPI site, a read-only MCP server and a fail-closed SQL guard for an English-question assistant.
  • Every AI decision in the platinum layer is recorded as a signed event in Entropa (Rust); keyless GitHub Actions publishing, CI with secret scanning, and a public build log. Spark and Scala code written by hand; tooling and infrastructure directed with AI and verified with tests.

MegaMart: an open data lakehouse, end to end (Spark, Iceberg, MLflow)

August 2026 – present · github.com/aimozart/MegaMart

  • Dirty data to trustworthy analytics for a fictional national retailer: a scale-factor-driven generator of deliberately dirty data (nulls, duplicates, orphan keys, mixed currencies, late events), then Bronze, Silver and Gold layers on Apache Iceberg over S3-compatible storage (MinIO locally, GCS in the cloud).
  • Silver: typing, deduplication with window functions, currency and timezone normalisation, orphan foreign keys quarantined (not dropped), sessionised clickstream with event-time watermarks, and slowly-changing dimensions (SCD Type 2) with Iceberg MERGE. Gold: business marts (revenue, customer 360, funnel, cohorts, fraud summary).
  • Machine learning on the Gold layer (churn, demand forecast, fraud) tracked in MLflow with a model registry, SHAP explanations and drift monitoring; Streamlit dashboards. Vendor-neutral open tooling, runs locally for $0 and on GCP unchanged.

Entropa: post-quantum, tamper-evident audit trail for AI decisions

May 2024 – Present · entropa.space · github.com/aimozart/entropa-public

  • ML-DSA-65 (NIST FIPS-204) signatures verified byte-for-byte against NIST's test vectors; a Certificate-Transparency-style Merkle log; only hashes stored.
  • Controls mapped from Google's HITRUST shared-responsibility matrix; secrets only in Secret Manager. Built and documented, not formally assessed.
  • gitleaks in pre-commit and CI (0 leaks across 348 commits) and a public incident log.

Skills

Forward-deployed AI (current project): controls evidence: post-quantum-signed, witnessed decision records; Cedar policy gates; evaluation sets with fixed bars, fresh held-out sets and attack tests; Rust; a standards retrieval system; test-first, with every stage forecast and then measured. The earlier simulated claims-triage case study is parked, not deleted.
Custom platform building (directed with AI, verified by tests): Rust (Entropa: a signed, serverless ledger of AI decisions on Cloud Run), Python and FastAPI, read-only MCP servers, a fail-closed SQL guard for AI-generated queries, GitHub Actions with keyless workload identity, Docker, Vertex AI (Gemini) for assistants.
Windows administration: Windows Server, Active Directory, Group Policy, Microsoft Intune, SCCM across all its functions (node and PXE server setups, OS imaging and task sequences, application deployment to users, policy, patching, baselines), Hyper-V, OS migrations (XP → 7 → 10, 7 → 11, an NT domain to Server 2012 R2).
Databases: MySQL (administration, phpMyAdmin, backups, restores, tuning), Lotus Notes/Domino, Databricks (Spark, Delta Lake).
Security & compliance: HITRUST CSF, HIPAA, CIS Benchmarks, least privilege, encryption and key management, endpoint DLP (LogRhythm Data Loss Defender), SIEM, EDR, PAM, vulnerability management, audit evidence.
FinOps: cost forecasting and unit economics, budgets and alerts tied to forecasts, price-performance sizing, tagging for cost allocation, forecast-vs-actual reconciliation; hardware and software licensing selection; vendor outreach and negotiation for the best deal.
Cloud & systems: AWS (certified: VPC, IAM, KMS, PrivateLink), GCP, Azure (Entra ID, RBAC, Policy, Key Vault, Defender for Cloud, Sentinel, Bicep), Linux (Ubuntu and RHEL), IBM i.
Scripting: PowerShell (since 2006), Bash, SQL, Python.

Certifications & recognition

Full Credly profile →

Professional experience

Systems Administrator — Independent Projects

Remote · December 2022 – Present

Hands-on Azure administration (Entra ID identity and access, Azure RBAC and Policy, Key Vault, Defender for Cloud and Sentinel, with Bicep and PowerShell) and data engineering (Spark, Delta Lake, Iceberg); building Entropa and a Rust standards-retrieval system in public (in progress); designed, built and operated Entropa end to end; database-migration and infrastructure-as-code projects on Azure and GCP; earned AWS SA-A and Databricks.

Senior IT Analyst & Security Administrator

Ultra Clean Technologies · June 2019 – August 2022

Windows-heavy: administered Microsoft Intune and SCCM across all its functions (node and PXE server setups, OS imaging, application deployment to users, policy, patching, baseline-configuration audits), Active Directory and Group Policy, with heavy role-based access control (RBAC), group administration and global (enterprise-wide) policy; part of the team that migrated a global fleet from Windows 7 to Windows 11; ran enterprise EDR; administered PAM; MySQL.

Desktop & Systems Support Engineer

Cognizant (TJ Maxx / HomeGoods account) · April 2018 – June 2019

Supported Active Directory and Hyper-V for roughly 3,400 TJX stores across the US and Canada (every North American banner, Sierra included), including the French-language stores, in a siloed environment; administered Microsoft Intune for the company's mobile devices across the enterprise (TJ Maxx and HomeGoods employees with company phones), including the C-level executives; delivered the enterprise-wide Software Center where entitled employees (mostly HQ staff and store managers) picked the applications they were permitted to install; administered Microsoft 365; administered Lotus Notes/Domino and databases for the account, and IBM i user access; C-level escalation support for down stores across the enterprise; restored 24/7 retail operations during outages.

Information Security & Compliance Analyst

ECS (acquired by CIOX Health, now Datavant) · April 2014 – October 2017

Heavy Windows administration: Active Directory, Group Policy, and SCCM (OS imaging, policy, node setups, PXE servers, application deployment to users); migrated Windows XP → 7 → 10, and a legacy Windows NT domain (about 300 users) and NT servers to Windows Server 2012 R2, policies and user accounts included. Co-authored the security-policy framework and drove remediation to official HITRUST certification; wrote the program's SOPs; owned the security stack; PHI chain of custody on FIPS 140-2 media; MySQL in a HIPAA environment.

Level 3 Linux Systems Engineer

Endurance International Group (now Newfold Digital) · July 2012 – February 2014

Linux web-hosting administration and occasional Windows Server hosting for customers who paid for it, plus MySQL and phpMyAdmin behind WordPress hosting, for 120+ enterprise accounts; server hardening and WordPress security on Ubuntu and RHEL. On the migration team for platform migrations across the company's hosting brands (Bluehost, HostGator, iPower and others): databases, WordPress sites, domains, DNS and Linux hosts, including moving acquired domains into larger platforms. Led the WordPress and hosting-account migrations.

Work arrangement

Education

Get in touch

Sent privately; your message goes straight to my inbox.