husthunterpy01/av-job-relevance-embedding

AV job relevance (frozen MiniLM + logistic probe)

Logistic regression head trained on Groq-labeled AV vs non-AV job postings. The encoder is the public sentence-transformers/all-MiniLM-L6-v2 checkpoint and is not fine-tuned.

Files

• relevance_model_embedding.joblib — sklearn LogisticRegression (class_weight=balanced)

Load

from scrapers.service.ml.embedding_classifier import EmbeddingRelevanceClassifier

clf = EmbeddingRelevanceClassifier.load("relevance_model_embedding.joblib")
# or, with no local file:
clf = EmbeddingRelevanceClassifier.load("/does/not/exist")  # downloads this repo

Same weights as python3 -m scrapers.utils.relevance_classifier_cli train --only-llm-labeled.