Evidence before adjectives
Metrics include their workload, environment, method, date, and reproduction command.
AI Agent · Platform · Full-stack Engineering
Engineering work designed to be verified.
I am Patrick Simonetti, working publicly as Patrick Dev. I design, build, and deploy AI agents and product systems across model/runtime integration, bounded tool execution, backend and web boundaries, infrastructure, and release evidence.
- Current focus
- Released AI systems and bounded local operations
- Operating model
- Evidence, contracts, automation, and recovery
- Location
- Rome, Italy
Selected engineering work
Four projects. Four distinct proof domains.
Each project owns one technical narrative. Shared standards apply across the portfolio, but featured status requires project-specific evidence.
v1.0.0 released · reproducible evidence
AI Prompt Optimizer
Select real files from the current project, add research when needed, and use your always-on home workstation from any IDE terminal.
Open project Public alpha · host execution explicitly boundedSysOps Agent
A local-first operations agent with fail-closed classification, bounded commands, verification predicates, recovery loops, and redacted audit trails.
Open project 2.0 release candidate · external gates pendingAutoBlog CMS
A durable editorial product with database sessions, workspace RBAC, immutable revisions, optimistic concurrency, leased jobs, and bounded media/AI operations.
Open project Main CI + security green · operator-managedAuthenticated Privacy Proxy
A containerized Mitmproxy–Privoxy–Tor chain with authenticated ingress, policy controls, DNSSEC-aware resolution, and portable host automation.
Open projectOperating principles
Professionalism is an evidence chain.
Architecture, product behavior, security, CI, deployment, and documentation must describe the same system.
Contracts before providers
Cloud, data, storage, identity, queue, and AI vendors terminate at tested adapters.
Failure is product behavior
Timeouts, retries, idempotency, migration, recovery, and rollback are designed explicitly.
DeployOps model
One artifact, progressively verified.
Portability comes from responsibility boundaries and contract tests, not duplicated application implementations for every provider.
- Source and tests
- Immutable artifact
- Provider adapters
- Ephemeral verification
- Observe and roll back
Deployment claims are tiered.
Reproducible, container-portable, platform-portable, provider-portable, and operationally verified are separate claims with separate evidence.
Technical discussion and opportunities
Start with the evidence.
Review the project cases, inspect the repositories, or contact me about engineering roles and product collaboration.