Skills are the primary index of the people directory. Titles vary by employer, bios vary by mood, but the skills list is where a person states what they can do in the fewest words. Searching it well is a small skill of its own.
Search the vocabulary people use, not the vocabulary reqs use
Profiles are self-described. People list react, not "frontend framework (3+ years)"; media-buying, not "performance marketing specialist". Before you search, translate:
- Job-post language ("rockstar", "ninja", "guru") into tool language (the actual stack).
- Formal discipline names ("information retrieval") into practitioner terms (
opensearch,vector-databases) -- then also try the formal term, because senior people often list both. - One broad term and one narrow term.
pythonfinds everyone;pythonplusairflowfinds pipeline engineers.
Combine, do not just add
A single skill is a superset. The search that finds your person is almost always a combination: skill plus location, or skill plus availability. "Rust, Berlin" and "Rust, open to freelance" are different searches with different best answers, and Parlel lets you run both in seconds. Start broad, look at who comes back, then narrow with the vocabulary the good results use -- the directory teaches you its own dialect if you read the first page before filtering further.
Skills are self-reported: calibrate
Anyone can list any skill, so treat the list as a claim and the rest of the profile as evidence. Strong corroboration, in order:
- The skill appears in a work-history bullet with an outcome attached.
- The skill appears in a public post demonstrating it.
- The employer is verified and the tenure is real.
A profile listing twelve skills with no evidence for any of them is a weaker signal than a profile listing three, each tied to a project. Recruiters sourcing at volume should make this calibration a checklist step, not a gut feel.
For machines: the same search as an API
Skill search is one of the eight MCP tools: search_people takes the same skill filters as the directory. If you are building screening, matching or analytics on top of Parlel, see the MCP guide -- and how Parlel search works for what each parameter matches.