I need an end-to-end anomaly detection system that keeps a constant watch on user behaviour and flags anything that looks suspicious in near real time. My goal is to go beyond simple thresholds and use machine-learning techniques—Autoencoders, Isolation Forests, One-Class SVMs, even CNN or LSTM variants if they prove the best fit—to uncover subtle, high-risk patterns that traditional rules miss. The workflow should cover data cleaning, feature engineering, model training, rigorous evaluation and, most importantly, seamless deployment. I want the finished model wrapped in a Streamlit dashboard that lets me review live traffic, visualise anomalies, and receive instant fraud alerts. Low-latency inference matters, so please plan to optimise the final model with TensorFlow Lite (or an equally lightweight alternative) to ensure quick decision making when we move to production. Here is what I expect as tangible deliverables: • Fully commented Python code for data processing, model development and inference • A Streamlit dashboard with real-time visualisations and an alerting mechanism • TFLite (or comparable) conversion scripts plus basic deployment instructions • A concise report detailing performance metrics, assumptions and next steps I will accept the work once the system achieves reliable precision/recall on a withheld validation set and demonstrates live alerting through the dashboard. When you place a bid, include a detailed project proposal that outlines your chosen modelling approach, key milestones and estimated timeline; bids without that will be skipped. Feel free to reference relevant past projects if they show you can build production-ready anomaly detection pipelines. I’m happy to answer clarifying questions quickly so we can move forward without delays.