Beak …🐤

Analyzes text, returns tags.

Name

Why Beak? Because beaks are good at sorting out the interesting bits in peck feeds.

AI tags

The technology tags include tags for how a text relates to AI. These tags are hierarchical — the more specific ones are always emitted together with ai, so filtering for ai catches everything and the others catch a sharper subset:

• ai — general AI awareness / chat-level use (the baseline): bare "AI", "umělá inteligence", ChatGPT, "AI nástroje", "AI-first", …
• agenticengineering — uses AI coding agents: named tools (Claude Code, Cursor, Codex, Copilot, …) and coding-agent phrasing. Bare "AI agent" is deliberately not mapped here — it usually means building agents, not using a coding assistant, so it only trips the general AI → ai rule, while "agentic architecture / systems" goes to buildingai.
• buildingai — builds AI / ML features: LLM, RAG, embeddings, prompt engineering, LangChain, vector databases, NLP, computer vision, and classic machine learning (machine learning, strojové učení, neural networks, TensorFlow, PyTorch, scikit-learn, MLOps, …).
• vibecoding — vibecoding as an attitude / way of working (implies agenticengineering): the many spellings of vibe/vajb + coding and the Czech verb forms, plus "vibe engineering" and "agentic engineering".

Individual products are intentionally not exposed as tags — ChatGPT and Claude are an equivalent skill, Codex and Claude Code are an equivalent skill, and the product names churn; what matters is the capability.

Two matching details worth knowing: the bare AI, ML, LLM, RAG and NLP abbreviations are matched case-sensitively (a case-insensitive ai would match unrelated lowercase substrings), and the tag boundaries come from a scrape and analysis of ~300 tech job postings on jobs.cz and startupjobs.cz.

Adding or editing rules

The matching rules live in src/jg/beak/mapping.toml as a list of [[rule]] entries — no Python needed. Each rule is a regex plus the tags it produces:

[[rule]]
pattern = 'claude code'
tags = ["ai", "agenticengineering"]

Conventions applied when the file is loaded:

• Matching is case-insensitive by default; set case_sensitive = true to turn it off (e.g. for AI, ML — a case-insensitive ai would match unrelated substrings).
• A plain space between words expands to \s+, so claude code also matches claude code and claude\ncode. Use explicit \s* / \s? / \x20 when you need something other than "one or more whitespace".
• The pattern is wrapped in \b…\b (whole-word match) automatically; set treat_as_word = false to match anywhere.
• Every tag is validated against the enums in tags.py on load, so a typo fails fast.

License

AGPL-3.0-only, copyright (c) 2024–2026 Jan Javorek, and contributors.