New Harbinger Publisher Assistant
Enterprise publishing and discovery multi-agent system
Designed and built a multi-agent publishing assistant that routes book discovery, ordering, rights and permissions, academic sales, author submissions, and media questions to specialized workflows grounded in verified publisher content.
Ask a real
publisher question.
Test book discovery or publishing support. The assistant retrieves grounded information and routes each request to the right specialist workflow.
- “Help me find an anxiety workbook for a teenager.”
- “Do you ship internationally?”
- “How do I request permission to reproduce material?”
Independent portfolio prototype. This demonstrates a proposed experience for NewHarbinger.com ↗. It is not currently deployed on the publisher’s production site.
Open demo in a new tabOne clear entry point for a complex publishing ecosystem.
New Harbinger serves readers, mental health professionals, academic buyers, authors, press, and international customers. Discovery and support tasks span a large catalog and several service paths. The product opportunity was to bring those journeys into one conversational interface without flattening their differences.
Product framing through working implementation.
I led the product concept, intent taxonomy, knowledge architecture, conversational UX, routing logic, retrieval strategy, guardrails, ChatKit interface, front-end integration, deployment, and testing.
Route first. Retrieve verified knowledge. Answer in context.
The classifier directs requests to book discovery, ordering and shipping, rights and permissions, bulk and academic sales, author submissions, press and media, general FAQ, or refusal and out-of-scope handling.
Useful answers need structure and boundaries.
- Ask only one or two clarifying questions before recommending
- Return three to five relevant books with title, author, rationale, and canonical URL
- Separate discovery from transactional and publishing-support intents
- Use boundary handling to avoid unsupported health or catalog claims
A reusable enterprise agent architecture.
- A deployed, interactive multi-agent prototype
- Grounded catalog retrieval across a 1,247-title knowledge base
- Structured routing for distinct customer and business needs
- A clear path to production catalog sync, analytics, human escalation, and publisher-site integration