Project overview
I helped architect the technical infrastructure required to operationalize AI for both customer support and internal production. The goal of this initiative was to replace unstructured, legacy documentation with a RAG-ready Strapi model to resolve critical hallucination issues in a help agent we were building, while simultaneously establishing a generative drafting pipeline to accelerate content creation.
Skills
Content modeling
Strapi CMS
RAG strategy
Prompt engineering
Multimodal AI workflows
Information architecture
The Problem
The organization wanted to move towards AI-assisted help workflows for a complex suite of retail data products. While the existing documentation was comprehensive and accurate for human users, the hosting platform lacked the strict metadata layer necessary for AI ingestion. Following a series of acquisitions, many products shared overlapping terminology and similar user interfaces. Without a way to explicitly map content to specific features and products, the AI agent suffered from "context collision"—it would retrieve accurate information, but often not apply it to the correct product or retailer.
The Outcome
I delivered a validated architectural blueprint and functional prototype centered on a RAG-optimized Strapi model. I atomized the documentation into discrete semantic units—separating concepts, tasks, and references—and established a rigid taxonomy to map content to its specific product and retailer. In testing, this structure solved the ambiguity issue, allowing the retrieval system to fetch precise content without conflating similar features or products. To increase writer productivity, I also designed an internal generative AI drafting workflow. I configured a custom AI agent with our specific style guide and schema requirements, enabling it to ingest UI screenshots and output structurally compliant drafts. This created a closed-loop system where AI accelerated content creation while the structured data ensured safe content delivery. This work established the necessary foundation and technical roadmap for the company to move forward with a reliable, hallucination-free automation strategy.