AANA MAX / OS29 → OS29
ENGINEERING PROFILE
Professional CV · Independent projects

Local AI tooling.
Real tests. Clear evidence.

Self-directed developer focused on local-first AI tooling, Windows automation, MCP integration and quality assurance. Builds practical Python and PowerShell workflows, then verifies behavior with bounded tests, live checks and documented limitations.

AI Tooling · Automation · QA & Reliability
Project-based engineering portfolio · ANA MAX OS29 / OS29
PUBLIC ANA MAX REPOSITORY ↗LINKEDIN PROFILE ↗
Local AI integration and Windows automation
Evidence-first QA, scoped access and readable reports
OS29 · operational intelligence

Professional summary

Concise and factual; no inflated title or unverified employment history.

01 / PROFILE

I work on the layer between AI-driven intent and local software: tool contracts, MCP/JSON-RPC integration, Windows-side automation, health checks, persistent diagnostic evidence and regression tests. I am interested in making complex agent workflows more observable and easier to verify—not just making a demo appear successful.

My projects combine implementation with test design and operational review. I separate declared, configured, running, tested, degraded and unknown states; document environment-dependent skips; preserve backups before risky changes; and keep a clear line between validated behavior and the roadmap.

Selected engineering work

Recorded results are labelled with their evidence date.

02 / PROJECTS
OS29 · CORE

Evidence-first operational intelligence

Recorded baseline · 19 Oct 2026

Built a local-first layer for normalized trace events, parent lineage, redaction before persistence, explicit target scope, incidents, mission capsules and capability-state evidence.

  • Defined a canonical event model with trace and parent identifiers instead of relying on unstructured logs alone.
  • Kept HTTP target scope fail-closed and Frida device/process allowlists empty by default.
  • Exposed a narrow MCP STDIO surface for observation, retrieval and evidence operations rather than arbitrary shell execution.
  • Recorded baseline: 37 local tests passed; a new repository CI run retests the published source independently.
OS29 · MEMORY & RETRIEVAL

Operational intelligence and evidence-linked memory

Local suite · 19 Oct 2026

Implemented persistent evidence storage with trace lineage, incidents, mission capsules, candidate-versus-validated experience states and bounded MCP STDIO tools.

  • 37 tests passed in the recorded OS29 suite; compile checks and the MCP STDIO initialize/list/memory-search smoke were reported as passing.
  • The reported checkpoint exposed 14 STDIO tools and indexed 375 local documents for SQLite FTS5/BM25 retrieval.
  • An end-to-end checkpoint verified three linked persisted events and retrieved both a validation event and a follow-up triage decision.
  • Known boundary: lexical retrieval is implemented; vector embeddings, reranking, a full model reasoning loop and the entire 30-module roadmap are not claimed as complete.
PORTFOLIO DEMO

Privacy-first browser profile

Static site

Created a browser-native live profile that personalizes the view using a display name chosen by the visitor, then renders the browser-exposed connection and device hints where supported. It makes no profile upload and does not pretend a static page can read Windows account names, full-system CPU/RAM or process lists.

Technical toolkit

Technologies grounded in the projects above.

03 / SKILLS
PythonLocal services, collectors, validation, scripts and unit tests.
PowerShell / WindowsLaunchers, process/service diagnostics, safe file workflows and system checks.
MCP / JSON-RPCTool catalogs, STDIO and HTTP integration, narrow client profiles.
QA & test evidencepytest, regression checks, smoke tests, syntax checks and explicit skip reporting.
SQLite / FTS5 / BM25Local persistence, event-linked records and lexical retrieval.
HTML / CSS / JavaScriptResponsive interfaces, browser APIs, input handling and static demos.
HTTP APIs & telemetryHealth endpoints, timeouts, bounded payloads and observable status.
Git / documentationBackups, scoped commits, repair reports, handoffs and working notes.
Privacy & scopeRedaction, local-only checks, explicit targets and clear trust boundaries.

Working principles

The habits I want visible to a hiring manager or collaborator.

04 / METHOD
A tool appearing in a catalog is not proof that it works. A passing test suite validates the tested surface, not the entire product. A dashboard status is not security certification. I try to make each of those boundaries visible in the code, the tests and the documentation.

LinkedIn “About” draft

Copy-ready summary for the profile page; edit to your preferred voice.

05 / INTRO
I build local-first AI tooling and Windows automation, with a strong focus on QA, reliability and evidence-driven engineering. My independent project work centres on ANA MAX OS29: MCP/JSON-RPC integration, Python tooling, scoped local diagnostics, persistent trace evidence, SQLite-based lexical retrieval and reproducible test suites. I care about the difference between a feature being declared, connected, tested and genuinely verified. I document limitations, preserve rollback paths and prefer measurable evidence over inflated claims. Core tools: Python, PowerShell, MCP/JSON-RPC, pytest, SQLite FTS5/BM25, HTTP APIs, HTML/CSS/JavaScript and Git. Portfolio: https://github.com/gyodragos-cell/ana-max-os29\nLinkedIn: https://www.linkedin.com/in/gyodragos/
View project evidence

Verification and scope notes

Important context for responsible, accurate presentation.

06 / NOTES
Test counts and index sizes above are recorded local snapshots dated 19 October 2026, not live counters. OS29 source and tests are published in this repository; local databases, runtime logs, secrets and approved-target identifiers are excluded. The examples describe independent project work, not employment by a named company or a certified commercial security product. The hosted portfolio is static and cannot launch local tools or verify a visitor's machine. Frida/MITM diagnostics require an explicitly authorized target and are not launched by the public page. No degree, employer, job dates or contact details have been invented.