JOBSELECT LABS / MODELS

Small models. Practical job intelligence.

Explore the AI models behind JobSelect's job-description analysis tools.

New modelSBERT · Semantic embeddings

Meet JobAnalyze SBERT

A sentence-transformer-based model path for job-description skill analysis. SBERT converts job text into semantic embeddings before classification, complementing the existing TF-IDF-based JobAnalyze 6k model with a representation designed to capture contextual meaning beyond exact keyword overlap.

Representation

Sentence embeddings

Semantic context from job-description text.

Pipeline

Embeddings + classifier

A separate model path alongside the TF-IDF baseline.

Best for exploring

Context-aware matching

Compare semantic signals with keyword-based predictions.

Model outputs depend on the trained artifacts and thresholds; semantic embeddings do not guarantee higher accuracy for every job description. Check the model documentation for evaluation details and current availability.

Explore models →Read documentation Source code ↗

TEXT CLASSIFICATIONv1.0

JobAnalyze 6k

A lightweight multi-label PyTorch model that predicts a fixed set of skills and keywords from a job description together with a provided role and job type.

Task

Multi-label classification

Framework

PyTorch

Features

TF-IDF

Hidden layer

32 units

Output

Skill probabilities

Vocabulary

48 labels

How it works

Job Description + Role + Job Type

Text Concatenation

TF-IDF

Linear Layer

ReLU

Dropout

Linear Layer

Sigmoid

Ranked Skills

max_features=150English stop wordsn-grams=(1,2)min_df=2hidden=32dropout=0.3

EXAMPLE OUTPUT

Top skills

APIs0.78

LangGraph0.78

VectorDB0.76

MCP0.75

LangChain0.74

RAG0.72

Illustrative frontend output only; no inference is performed here.

Evaluation snapshot

Micro-F1

0.624

Macro-F1

0.420

Baseline Micro-F1

0.538

Baseline Macro-F1

0.152

SAMPLE INFERENCE TEST

93.34% recall

14 / 15 target skills. Sample result, not universal accuracy.

Training data & label space

The documented published dataset contains 609 rows.

roletypejob_descyearqualificationexperiencetech_skillssoft_skills

View Dataset

Know the limits

Fixed vocabulary. Predictions are constrained to the documented skill vocabulary.

Dataset coverage. Performance depends on labelled-data coverage and quality.

Not a hiring predictor. The model analyzes job-skill signals; it does not determine whether a candidate will be hired.

Scores are model outputs. They should not automatically be treated as calibrated real-world probabilities.

Want to see the model in action?

Open Job Analyzer