Contact center representatives manually create thousands of service requests under real-world time constraints. Built a multi-agent AI system that classifies, generates, and routes requests with significantly higher accuracy and speed. Designed to scale across multiple business groups with measurable impact on quality and handling time.

Large language models lack the company-specific knowledge, operational expertise, and conversational context that experienced contact center representatives build over years. Built a multi-turn Conversational RAG pipeline with structured dialog management paired with a domain-adapted LLM through continuous pre-training and fine-tuning. Evaluated through an expert pilot study with contact center representatives, demonstrating strong retrieval quality, response acceptability, and human ratings.

Manually extracting information from complex, varied 401-K business documents is a months-long bottleneck that limits how many clients the business can onboard at any given time. Built an end-to-end multi-modal document extraction pipeline combining custom layout detection, hierarchical document parsing, fine-tuned retrieval models, and generative LLMs to automatically extract a large number of fields from scanned and digital documents. Deployed in production, significantly reducing manual effort and enabling the business to scale client onboarding.