First-party project / Applied AI
A useful conversation.
Considered engineering.
Ask ZuBOT helps visitors explore my work, understand my services and take the next step. A focused AI feature, built into the portfolio it represents.
- My role
- Design & full-stack implementation
- The stack
- Laravel · Vue · OpenRouter · Redis
- The scope
- Portfolio guidance & enquiry handoff
The problem
Make the next step
feel clearer.
A portfolio contains many projects, services and details. Visitors should be able to ask a direct question and find a relevant route through that information, then contact me when they want to discuss a role or a project.
I built the interface, server-side conversation flow, knowledge integration and email handoff as one connected experience.
What could we build together?
Connected to a model.
Request limits · Conversation context · Email handoff
The implementation
Useful AI takes
more than a prompt.
The decisions behind the conversation, from context to a human response.
Give it relevant context.
Curated Markdown packs describe my background, products and services. A fuller prompt adds portfolio and testimonial context; a compact prompt keeps simpler exchanges lean.
Connect the whole flow.
A Vue interface talks to Laravel conversation endpoints. The server calls the model through OpenRouter and retains recent conversation context. Simple routing rules select the model and prompt tier for different enquiries.
Put limits around access.
Server-side message-length checks and Redis-backed per-minute and per-hour request limits help manage misuse and model costs. API credentials stay on the server.
Leave a path to a person.
When an enquiry needs my attention, the conversation can request an email address and send a handoff with its transcript. Visitors can also use the contact form directly.
Engineering judgement
Keep the scope clear.
Keep a person in reach.
ZuBOT is a knowledge-guided assistant. It uses supplied context rather than a vector-search retrieval pipeline, and it does not act as a general-purpose autonomous agent. That focused scope fits the job: helping people navigate this portfolio.
Model responses can still be inaccurate. Request limits and instructions reduce some risks; they do not guarantee correct answers or prevent every form of misuse. Important project details, commitments and estimates belong in a direct conversation with me.
The same approach shapes my delivery work: review authentication, authorisation, data handling and failure paths, including changes made with AI assistance.
From an idea to a useful feature
Have an AI integration
in mind?
Tell me what people need to accomplish, where the information lives and when a human should take over.