Senior AI & Software Systems Engineer — Agent Evaluation, Runtime Governance, Safety Engineering & Native Platforms
- 🛡️ AI Agent Evaluation & Runtime Governance: Designing and operating production safeguards for tool-using, multi-step LLM agents that execute real-world actions (
send_message,update_record,prepare_decision). Enforcing accountable ownership, human/campaign approval workflows, data provenance restrictions, deterministic pre-execution checks, and automatic quality-based suspension (kill switches) in code. - 📊 Applied Statistics & Go-Live Release Gates: Implementing statistically rigorous evaluation frameworks in Python and TypeScript—including Cochran finite-population review sample sizing, Wilson Score 95%/99% error-rate confidence bounds, and two-proportion Z-test regression detection across model or instruction changes.
- ⚡ Frontier AI-Native Engineering & LLM Workflows: Advanced mastery across modern agentic engineering platforms—including Google Antigravity, OpenAI Codex, and Claude Code—combined with hands-on LLM engineering using the latest model generations such as OpenAI Astra 6 (and GPT/o-series), Google Gemini 4 Argo, and Anthropic Claude. Grounded in senior software engineering foundations, I leverage these technologies to dramatically accelerate delivery velocity while strictly enforcing architectural best practices, deterministic verification, and end-to-end project security.
- 🔌 Model Observability & Provider Independence: Architecting vendor-agnostic model abstractions (
OpenAI Astra 6,Google Gemini 4 Argo,Anthropic Claude, and on-device models) with immutable audit ledgers recording model ID, instruction version hash, data provenance, and deterministic quality checks on every invocation. - 📱 End-to-End Native & Cloud Systems: Translating written governance rules, ontologies, and acceptance matrices (
MATRIZ-PARIDAD.md,ADR-0001) into executable validation suites across TypeScript/Node.js, Python, PostgreSQL RLS, iOS (Swift 6 / SwiftUI), and Android (Kotlin 2.2 / Jetpack Compose).
"Modern agentic AI tools multiply speed—but senior engineering discipline ensures that speed never compromises reliability, maintainability, or security."
| Dimension | Platforms, Frontier Models & Engineering Discipline |
|---|---|
| 🤖 Agentic IDEs & Autonomous Coding Systems | Expert command of Google Antigravity, OpenAI Codex, and Claude Code for multi-agent orchestration, multi-repo refactoring, automated test synthesis, and spec-driven verification. |
| 🧠 Frontier LLM Development Stack | Building, evaluating, and orchestrating multi-step workflows with OpenAI Astra 6 (plus GPT / o-series), Google Gemini 4 Argo (plus Gemini Pro / Flash), and Anthropic Claude, abstracted behind provider-independent interfaces. |
| 🔒 Integrity, Best Practices & Security Guardrails | Every AI-accelerated feature is gated by Architecture Decision Records (ADR-0001), deterministic test suites (14,449+ checks in Myaow), statistical regression gates, zero-secret repositories, HPKE RFC 9180 E2EE, and PostgreSQL Row-Level Security (RLS). |
| System / Repository | Domain & Stack | Governance, Evaluation & Architecture Highlights | Links |
|---|---|---|---|
🛡️ agent-governance-runtime |
TypeScript · Python · Applied Statistics | Governed Multi-Step Agent Runtime & Evaluation Gate: Cochran review sample sizing, Wilson 95% error upper-bound gating, model regression Z-test detector, HITL & campaign-level approvals, data classification guards, immutable model call provenance ledger (OpenAI Astra 6, Gemini 4 Argo, Claude), and automatic kill switches. Includes ADR-0001. |
Repo · Live Lab |
🐾 Myaow-iOS |
Swift 6 · CryptoKit · WebRTC · On-Device AI | E2EE Communication & Collaborative Workspace Ecosystem: 14,449+ deterministic pre-release validation checks (validate-critical.sh), governed proposal-to-task approval workflows (ProjectWorkspace), fail-closed on-device AI content moderation, HPKE RFC 9180 Auth v2, Hybrid PTT Radio & 23 procedural H.264 wallpapers. |
Repo · Case Study |
🤖 Myaow-Android |
Kotlin 2.2 · Jetpack Compose · Keystore | Hardware-Backed Native Client & Executable Parity Matrix: Converts 110+ written acceptance scenarios (MATRIZ-PARIDAD.md) and ADR-0001 into automated cross-platform Swift ⇄ Kotlin cryptographic fixtures and explicit digest-confirmed device linking (DeviceLinkPrototype). |
Repo |
🏙️ aylesvoy-smart-directory |
Serverless · Vision AI · Firebase · iOS | Fail-Closed AI Content Moderation & Cost Governance: Automated image safety evaluation pipeline (Vision SafeSearch) with strict fail-closed rejection when confidence thresholds or upstream availability fail, plus 90%+ cloud cost optimization. | Repo |
🏛️ Aylesva |
Full-Stack · Shopify OS 2.0 · REST/GraphQL | Binational digital services & B2B corporate marketplace operating across Mexico & USA with modular architecture and automated validation pipelines. | Live · Repo |
🗳️ Pascuali |
SwiftUI · MapKit · GeoJSON Analytics | Geospatial field-operations iOS platform rendering 105 polygon sections with deterministic Ray-Casting point-in-polygon validation and proportional target modeling. | Repo |
🍇 Raíz de Plata & Mundo Country |
Shopify Theme 2.0 · Liquid · JS | High-conversion e-commerce storefronts engineered with modular section schemas, strict input validation, and Core Web Vitals optimization. | Raíz de Plata · Mundo Country |
┌──────────────────────────────────────────┬──────────────────────────────────────────┐
│ AI Agent Governance & Safety Engineering │ Evaluation Metrics & Applied Statistics │
├──────────────────────────────────────────┼──────────────────────────────────────────┤
│ • Tool-Using Multi-Step Agent Runtimes │ • Golden Datasets & Go-Live Release Gates│
│ • Accountable Action Ownership in Code │ • Cochran Finite-Population Sample Sizing│
│ • Single-Action & Campaign Approvals │ • Wilson Score & Clopper-Pearson Bounds │
│ • Deterministic Pre-Execution Safeguards │ • Two-Proportion Z-Test Regression Gates │
│ • Automatic Quality Suspension / KillSw │ • Model & Instruction Version Comparison │
│ • Model Observability & Data Provenance │ • Executable Specs & ADRs Before Coding │
│ • Provider-Independent LLM Abstraction │ • Antigravity · OpenAI Codex · ClaudeCode│
│ • OpenAI Astra 6 & Google Gemini 4 Argo │ • Fail-Closed Safety & PII/Scope Guards │
└──────────────────────────────────────────┴──────────────────────────────────────────┘
Frontier AI Tooling, LLMs & Agent Orchestration:
AI Systems, Backend & Languages:
Native Mobile, Cryptography & Web:
-10B981?style=for-the-badge&logo=checkmarx&logoColor=white){kind=icon}
