v1.0.0

First tagged release of the Real-Time Fake Job Detector: a streaming pipeline that scores job postings as they arrive. Kafka feeds a Spark ML job, results land in PostgreSQL, a FastAPI service exposes them, and a Next.js dashboard shows them live.

Model (from docs/metrics.json)

Trained on the Kaggle Fake Job Postings dataset (17,880 postings, 4.8% fraudulent), evaluated on a held-out stratified test set of 3,576 rows with a validation-tuned decision threshold of 0.28:

Metric	Value
F1 (primary)	75.6%
Precision	82.3%
Recall	70.0%
Accuracy	97.8%

Accuracy is not the headline metric: always predicting "real" already scores 95.2% on this data. Confusion matrix: TN 3377, FP 26, FN 52, TP 121.

Highlights

• Imbalance-aware training: class weighting, stratified splits, and a threshold tuned on a validation slice carved from the training data
• Live dashboard with fraud alerts, trends, investigation and system monitoring views
• Docker Compose setup for the full stack

Changes in this release

• CI now passes: Python lint (black, isort, flake8 with a committed config), Python tests, and the dashboard type-check and build
• Dashboard CI job switched from npm to pnpm to match the committed lockfile, and the two dependency build scripts that pnpm requires approval for are allowlisted
• Removed three broken, unreachable components left over from a marketing template (campaign feed, asset grid, assets page)