Anomaly Review & Action Console (Retool + AI)

Title: Anomaly Review & Action Console — AI-Powered Fraud Detection Dashboard Summary/Tagline: Built a full-stack internal tool that automatically detects suspicious transactions using machine learning and lets teams review, act, and track them — all in real time. The Problem

Businesses handling high transaction volumes struggle to catch suspicious or anomalous orders quickly. Manual review is slow, inconsistent, and doesn't scale — by the time a problem is spotted, the damage is often already done. Teams need a way to automatically flag risky activity, understand why it was flagged, and take action without switching between five different tools. The Approach

I designed and built a complete internal tool using Retool as the frontend, connected to a live PostgreSQL database (hosted on Neon), with a Python-based backend for the machine learning layer.

Anomaly Detection: Used Isolation Forest to automatically identify unusual transaction patterns across the dataset — no manual rule-writing required.

AI Explanations: Integrated Groq running Llama 3.3 to generate plain-English explanations for why each transaction was flagged, so non-technical reviewers can understand the reasoning instantly.

Real-Time Actions: Built action buttons directly into the review queue — approve, escalate, or dismiss flagged orders on the spot, with every action logged to an audit trail.

Regional & Time-Series Insights: Added charts breaking down flagged orders by region and by time, so teams can spot patterns, not just individual incidents.

Resolved a DNS resolution issue connecting to the Neon PostgreSQL endpoint by hardcoding the host address.

Fixed schema caching errors that were causing stale data views.

Solved a hanging update loop by switching to chunked batch writes, making the system stable under real data loads.

A fully functional, end-to-end internal tool — not a prototype — that takes raw transaction data all the way to actionable, explained insights. It demonstrates the complete pipeline: database → ML detection → AI explanation → human action → audit log.

Tech Stack Retool · Python · PostgreSQL (Neon) · Isolation Forest (scikit-learn) · Groq (Llama 3.3) · SQL