← Back to Home

LA Department of Water and Power

AI Chatbot & Revenue Recovery System

Timeline: Summer 2025 (June–August) Role: Software Engineering Intern Location: Los Angeles, CA
LADWP AI Chatbot Interface

Overview

Designed and deployed an AI-powered chatbot using SharePoint + OpenAI RAG to retrieve LADWP technical standards, achieving ~80% faster lookup speed and higher accuracy than existing tools. Also developed a revenue recovery system to detect unmetered Accessory Dwelling Units (ADUs) from 80K+ work requests, flagging 206 high-risk cases.

Impact

~80%
Faster Lookup Speed

Compared to manual search through technical documents

250+
Staff Adoption

Projected usage across Power division

$370K
Annual Revenue Recovery

With multi-million dollar scaling potential

206
High-Risk Cases Flagged

From 80,000+ work requests analyzed

Tech Stack

Python OpenAI API (GPT-4) ChromaDB Tesseract OCR Microsoft Copilot Studio SharePoint Microsoft 365 Excel/SQL

Key Features

AI-Powered Technical Standards Retrieval

Built RAG (Retrieval-Augmented Generation) system to help LADWP engineers quickly find technical standards and documentation, replacing slow manual search processes.

Revenue Recovery System

Developed automated detection system for unmetered Accessory Dwelling Units (ADUs), analyzing work request data to identify properties that should be billed for electricity but aren't currently.

OCR Integration

Implemented Tesseract OCR to extract text from scanned technical documents, enabling comprehensive search across both digital and legacy paper-based standards.

SharePoint Integration

Seamlessly integrated with existing LADWP infrastructure through SharePoint and Microsoft 365, ensuring easy adoption without disrupting current workflows.

Challenges & Solutions

Challenge: Legacy Document Formats

Solution: Implemented multi-format data extraction pipeline using Tesseract OCR for scanned documents and custom parsers for various digital formats, ensuring comprehensive coverage of LADWP's technical library.

Challenge: Massive Work Request Dataset

Solution: Developed efficient data processing pipeline to analyze 80,000+ work requests, using pattern matching and machine learning techniques to identify high-risk ADU cases with high precision.

Challenge: Enterprise Security & Compliance

Solution: Worked within LADWP's security requirements by leveraging Microsoft's existing infrastructure (SharePoint, Copilot Studio) and ensuring all data remained within the organization's secure environment.

Feedback

"This tool could scale department-wide, even to Water [division]."

— LADWP Staff Feedback

Tags

AI/ML Python RAG Municipal Infrastructure Revenue Recovery Enterprise Software