2026 — present · Coming soon
A working toolkit for resilience testing and load-tool comparison. Pure Python 3 standard library, runs on a laptop, no cloud required. Built alongside three Russian-language study courses that I wrote as my own structured notes.
A headless CI agent for triaging failed tests: it parses MSTest and Selenium .trx reports, then groups and prioritises failures through an LLM pipeline. Local Ollama as the baseline, escalation to the Anthropic API only for the hard classes — a controlled quality / cost / latency tradeoff.
A desktop tool with a local knowledge base: RAG on ChromaDB and Ollama, an MCP server exposing 39 tools to Claude Code, and A/B testing of prompts with measurable quality before and after.