Skip to content
Back to portfolio

AI-assisted working method

AI applied to networks, systems, and automation

I use generative AI, code agents, and local models as technical copilots to speed up diagnosis, documentation, automation, and change review in infrastructure environments. I work with scoped context, protected data, local validation, Git diffs, and human review before applying changes.

01

Context

Repo, goal, constraints, sensitive data, and limits on what must not leave.

02

Plan

Hypothesis, scope, affected files, validation criteria, and rollback.

03

Execution

The right agent or model for the risk: cloud, local, CLI, or manual review.

04

Wrap-up

Reviewed diff, tests, documentation, traceable commit, and reusable learning.

Operational impact

  • Faster initial diagnosis of recurring incidents through assisted analysis of logs, metrics, and runbooks.
  • More consistent operational documentation for infrastructure procedures.
  • Lower risk in production changes by requiring diffs, local validation, and human review.
  • Better signal classification before alerting or blocking, reducing false positives.

Infrastructure diagnosis

Assisted analysis of logs, metrics, errors, configurations, and runbooks to reach a verifiable technical hypothesis sooner.

Per-repo code agents

I use Codex, Claude, OpenCode, and Gemini with per-project rules to implement small changes, review diffs, and uphold technical standards.

Local LLMs and sensitive data

I use Ollama and Open WebUI for workflows where operational context should stay within my own infrastructure.

Automation and scripts

I speed up CLI tools, parsers, API integrations, operations scripts, and executable documentation without giving up human review.

Operational security

I use LLMs as a support layer to classify context, explain signals, and reduce false positives; never as the sole blocking authority.

Quality, control, and traceability

Every assisted change ends with human review, lint, build, tests when applicable, Git diff, documentation, and a traceable commit.

Evidence in projects

  • CentinelaLocal Ollama inside the threat intel pipeline to classify context and reduce false positives before alerting or blocking.
  • Infra docs / runbooksOllama/Open WebUI procedures and Elastic/Proxmox log analysis with a local model and operational output.
  • PortfolioNext.js, CONTENT/AGENTS rules, anonymized screenshots, local validation, and a Codex-assisted commit workflow.
  • Per-repo rulesAGENTS.md, CLAUDE.md, RTK.md, and project-specific prompts to give code agents stable context.
  • Dashboards and CLIsAssisted review for Python/JavaScript tools with APIs, caching, reports, local dashboards, and operations workflows.

Controls I apply

  • Never publish secrets, credentials, or raw production data.
  • Use a local LLM when operational context must not leave.
  • Work with scope limited to a repo, file, or task.
  • Validate with lint, build, tests, or a reproducible manual check.
  • Review the Git diff before accepting changes.
  • Document decisions, evidence, and rollback when applicable.

Principle: AI proposes, the operator validates.