Built things · Построенные вещи
Working systems, each mapped to a line in the job description. Repos are public.
Python pipeline that parses GLOSS Russian lesson XML into an Anki deck: 7,117 vocabulary entries, 2,159 sentence cards, 6,065 audio files, 9 topics, ILR levels 1–3+. XXE-hardened parsing, plus a DLPT-style listening practice app.
In use · 7,117 cardsOracle ARM free tier running Jellyfin, Kavita, and Shelfmark behind a Cloudflare Zero Trust tunnel. Default-deny firewall, fail2ban, key-only SSH, unattended upgrades — zero inbound exposure beyond SSH. Full manifest and hardening runbook in the repo.
In productionSystemd timers watching market hours — premarket, intraday, end-of-day — with escalating drop alerts, plus an agent-scheduled monitoring fleet alongside. User-level, self-documenting, one-command install.
In useThe site you're reading: a custom EN/RU i18n system with a live language toggle, a design system on PT Serif/Sans and IBM Plex Mono, shipped from a public repo through GitHub Pages and Cloudflare DNS. Built end-to-end with AI-assisted workflows.
LiveStandard-library-only NLP for Russian material: corpus ingestion, tokenization, frequency and KWIC concordance extraction, self-contained HTML report. Runs anywhere Python 3.8+ does — the constraint real analysis tools face.
In useSeeded synthetic networks (planted communities, preferential, small-world) → degree distribution, edge-betweenness bridges, label-propagation communities → self-contained SVG report. Bridges render in red to train the high-value-node instinct.
In useDocumented capture-to-product methodology with an executable collector: query-syntax library, a 5-axis source-evaluation rubric scored into every provenance sidecar, and rate-limited, passive-first collection.
In useScheduled collection feeding a templated intelligence product — collection → graph analysis → report, automated end to end. Wired to the collection workflow and the graph lab; systemd timer included.
In use