Tagging API: Programmatic Medium and Large, more accurate and faster

Tagging Algorithms on the Programmatic Medium and Programmatic Large models are updated. They return more accurate tags than Programmatic Small, Programmatic Medium processes each text 2.3× faster, and the descriptions you write in the workspace for your tags, your taxonomy and your algorithm now guide the result.

Action required

Now: none. POST /text/tagging keeps its parameters and its response. Programmatic Small and the dynamic tagger are unchanged.
Later: none. Nothing in this update is deprecated.

What changed

Programmatic Medium and Large process each text faster: 2.3× faster for Medium, 33% less time for Large.
Programmatic Medium and Large return more accurate tags than Programmatic Small, making better use of each tag's description.
With Programmatic Medium and Large, the description of the Tagging Algorithm and the description of its taxonomy now guide which tags are returned. Use them to say what to tag, for example "the contract the candidate is looking for, not a past one".
With Programmatic Medium and Large, predictions are relevance scores between 0 and 1. Tags judged irrelevant score 0 and come after the others. If you filter on a score threshold, review it.
With Programmatic Medium and Large, the taxonomy's default tag (Other) can be returned first when no tag fits the text.

Performance

Model time per text (median)	Before	Now
Programmatic Medium	1.71 s	0.74 s (2.3× faster)
Programmatic Large	2.69 s	1.78 s (33% less time)

On our internal benchmark, Programmatic Medium makes a median of 37% fewer top-1 errors than Programmatic Small (between 14% and 99% depending on the taxonomy). Writing the algorithm and taxonomy descriptions cuts errors by a further median of 5%, up to 50%.

Example

curl -X POST https://api.hrflow.ai/v1/text/tagging \
  -H "X-API-KEY: <your_api_key>" \
  -H "X-USER-EMAIL: <your_email>" \
  -H "Content-Type: application/json" \
  -d '{"algorithm_key": "<your_algorithm_key>", "texts": ["<a job offer>"], "top_n": 2}'

{
  "code": 200,
  "message": "Tagging predictions ended successfully in 0.41 seconds!",
  "data": [
    {
      "ids": ["data_engineering", "cloud_infrastructure"],
      "tags": ["Data Engineering", "Cloud & Infrastructure"],
      "predictions": [0.91, 0.62]
    }
  ]
}

Action required

What changed

Performance

Example

Links