RAG Support Assistant at CVS Health
Self-service RAG assistant over ticket history and internal documentation. The assistant deflected 16% of tickets and cut resolution time by 20% while fitting CVS enterprise platform constraints.
What this project proves
Internal AI platform builder
Self-service RAG assistant over ticket history and internal docs that deflected 16% of tickets and cut resolution time by 20%.
Core challenge
Make internal support knowledge self-serve without losing governance or deployment rigor.
Evaluation lens
RAG architecture, platform integration, and enterprise deployment.
A support assistant that reduced manual support load and improved troubleshooting speed.
Overview
At CVS Health, I built a self-service RAG assistant over ticket history and internal documentation. The goal was to make recurring troubleshooting knowledge available directly to engineering teams instead of routing every question through manual support.
The current resume reports two measured outcomes: the assistant deflected 16% of tickets and reduced resolution time by 20%. It also fits the broader CVS platform environment: GCP, Terraform, ArgoCD, governance constraints, and internal documentation workflows.
What I Owned
- Built the retrieval and generation workflow over ticket history and internal docs.
- Shaped the assistant around self-service troubleshooting rather than generic chat.
- Integrated the workflow with enterprise deployment and governance expectations.
- Used the system to reduce repeat manual support work for engineering teams.
Hard Problems Solved
- Ground answers in internal evidence: support answers needed to be based on retrieved historical tickets and documentation.
- Reduce support load without hiding risk: the assistant had to resolve common issues while preserving a path for unresolved cases.
- Fit enterprise platform constraints: deployment and operations needed to align with CVS cloud and infrastructure practices.
Impact
- Deflected 16% of tickets.
- Cut resolution time by 20%.
- Improved knowledge discoverability across recurring troubleshooting workflows.
Why It Matters
This project shows RAG as an operational support system rather than a demo: retrieval quality, deployment fit, and measurable support outcomes all mattered.
Tech Stack
- AI/ML: Retrieval-Augmented Generation, LLM answer generation
- Infrastructure: GCP, Terraform, ArgoCD
- Backend: Python
- Data: Ticket history, internal documentation, runbooks