WordPress + Sitefinity
Multi-tenant AI chatbot platform
RAG-based chatbot product as primary and sole developer — 6–7 client sites, 30,000+ documents indexed, streaming that cut perceived latency by 20–30 seconds.
Brief
Context
A RAG-based AI chatbot product, architected and built as the primary and sole developer, for agency client sites running on two CMS platforms: WordPress and Sitefinity.
Build
What I built
Nightly content indexing writes to a vector store through WordPress REST APIs or Sitefinity APIs, with custom logic per content type. Events, for example, auto-index only future dates and auto-delete past ones during the index cycle.
The most complex deployment combined AI-enhanced site search with separate public and private chatbots. Segmented vector-store collections prevented duplicate indexing while honoring field-level visibility differences between public and private audiences.
WordPress-side proxy endpoints keep API keys off the client: the browser talks to WordPress; WordPress signs and forwards to AWS.
An on-site private dashboard lets clients monitor conversations and user feedback.
For the Sitefinity (.NET) handoff, a minified Vue bundle is configurable entirely through data attributes on the mount element — no rebuild per site.
Technology
Stack
- Vue frontend delivered via WordPress plugin or standalone CDN script
- Node.js on AWS Lambda
- DynamoDB
- API Gateway
- EventBridge (nightly indexing)
- Secrets Manager
- Serverless Framework IaC with per-client stages
Results
Outcomes
- Deployed to 6–7 client sites across WordPress and Sitefinity.
- 30,000+ documents indexed on the largest deployment.
- Streaming responses cut perceived latency by 20–30 seconds on long answers — first tokens arrive in 3–4 seconds instead of waiting for full generation.