AI search
Write what you are looking for in plain language. The answer streams in with the tool steps visible, and the candidate cards land underneath with level and evidence.
The product
Week29, view by view.
A recruiter panel that opens from Teamtailor. Here is what it does, and what happens in the engine when you press Match.
The panel
Four views from the workflow.
The panel does one thing: it surfaces the strongest candidates you already have.
Job matching
Import an ad from Teamtailor. The text of the ad becomes the role's requirement text, and the whole database is ranked against it.
Results and comparison
A ranked list with levels and requirement coverage. Select several candidates and set them against each other on relevance, experience, requirements, location and availability.
The detail sheet
The requirements, most important first, with a status for each and evidence where the candidate's text supports the requirement, the candidate's own material, and the actions to pass along or save.
Ask in plain language and learn who meets the requirements, and why.
The AI search in Week29 with a question in plain language, an answer with evidence and candidate cards with match levelsThe engine
What happens when you press Match.
Four steps, and each of them can be explained to an auditor.
The database is loaded
Candidates are pulled from Teamtailor with CV and free text. Changes are fetched within a day, so the database stays current without anyone syncing by hand.
The text is made searchable by meaning
Each candidate's material is split into passages at sentence boundaries and stored as vectors, so that the meaning stays together.
The whole database is ranked
The role's requirement text is compared against every candidate in the database, by meaning and by word matches. The order is deterministic: the same input gives the same result.
The evidence is produced
The key terms of the role are split into those that appear verbatim and those that are missing. For the missing ones we look in a fixed list of phrasings that mean the same. That step is entirely deterministic, with no language model involved.
Guiding principles
Four rules in the code.
Levels, not numbers
Matching is shown as Strong, Good or Possible, never as a percentage or a score. The same goes for what is written back to Teamtailor: a level and a tag, never a number.
Write actions are gated in code
The AI can propose passing candidates along, but never do it. The server checks every candidate against the database, not against the chat history.
A quote when the text supports the requirement
When the candidate's text supports a requirement, the sentence is shown as a verbatim quote from the candidate's own material. The reasoning next to it is built from fixed rules, not by a language model.
Instructions in a CV are stopped
A CV can contain instructions aimed at an AI. We sanitise all candidate text and flag profiles that look manipulated for human review.
Teamtailor
On top of the ATS you already chose.
Week29 replaces nothing. Your workflows stay in Teamtailor, and the result lands where you already work.
- The match level is written back to the candidate card through your trigger in Teamtailor. From the panel, the tag is written.
- The panel opens from Teamtailor and links back to the candidate.
- Changes in Teamtailor follow within a day, with nobody syncing by hand.
Built as a Teamtailor integration. Activation happens together with you.
Demo
See it run against a real database.
We show the ranking, the evidence and the controls, and answer the questions about data and regulation. It takes thirty minutes.