ThreatOps AI

Threat Ops AI is an AI-powered fake job offer detection and recruitment fraud intelligence system built entirely on Databricks, designed to protect India's 5 crore migrant and daily wage workers from the organized scam networks that cost them over ₹10,000 crore in losses every year. The system begins with a structured data ingestion pipeline that pulls labelled job posting data into a Bronze-Silver-Gold medallion architecture on Delta Lake, where raw records are progressively cleaned, normalized, and feature-engineered into a governed, fully traceable dataset managed through Unity Catalog — giving the project enterprise-grade data lineage from the very first byte. On top of this foundation, a TF-IDF feature extraction layer transforms job post text into high-dimensional vectors capturing unigrams, bigrams, and trigrams across job titles, descriptions, requirements, and company profiles, which are then fed into an XGBoost gradient boosting classifier trained on over 17,000 real and fraudulent job postings, with every experiment, metric, and model version tracked and registered in MLflow's Model Registry for full reproducibility. The classification model alone is not enough — so Threat Ops AI pairs it with a Mosaic AI large language model endpoint that receives both the fraud probability score and the original post text, and returns a structured explanation identifying the exact three phrases or patterns that triggered the alert, translated into plain language that requires no technical literacy to understand. The entire dual-model inference pipeline is wrapped in a Databricks App with a clean paste-and-analyse interface, and complemented by an AI/BI Genie Space where investigators, NGOs, or government officers can query the underlying Delta Lake tables using natural language — asking questions like "which job categories are most targeted in Karnataka?" or "how has fraud language evolved month over month?" — and receive live chart responses without writing a single line of SQL. With a 94% F1 score on the fraud class, a full data governance story through Unity Catalog, and a clear roadmap toward a zero-friction WhatsApp bot that requires no app download or digital literacy, Threat Ops AI transforms Databricks from a data engineering platform into a frontline instrument of economic protection — one that can tell a vulnerable worker the truth about a job offer in under 3 seconds, before their savings are gone.

Here's the github : - https://github.com/Sai-Emani25/ThreatOps-AI