AI Projects
Two working pieces built with AI. Each takes on one specific job, shows where its answers come from or where it stops, and leaves the final call with a person.
Interactive example · runs in your browser
Visitor Answer Desk
A desk for answering visitor questions about scenic train trips. Choose a question and it shows the source lines that apply, a prepared reply you can edit, and anything the sources leave open.
- Pick a visitor question from the list.
- Check the reply against the source lines beside it.
- Edit it, mark it reviewed and copy it into your own email.
Two of Mountain Rail's pages describe food at the Whittaker Station stop differently. Open the lunch question to see a reply that says only what both pages support.
Built on Mountain Rail WV's public pages, checked Oct 1, 2026, with fictional questions. The code, reference summaries and sample replies were drafted with an AI model, and every reference point links to the public page it came from. No AI runs on the page itself, and nothing is sent. Independent; not affiliated with Mountain Rail WV.
Build Week 2026 · case study
Calqen
Calqen helps a person decide what to wear from clothes they already own. Give it an occasion, the conditions, a style direction and a wardrobe, and it returns three complete looks with reasons. One look can then be refined while the source pieces stay visible.
Decisions the model didn't make
- What “better” meant for this person in this moment.
- That a refinement changes the piece that was asked about and keeps the rest.
- That a fallback is labeled honestly when a live result isn't available.
Built for OpenAI Build Week 2026, with GPT-5.6 working inside a bounded product flow; the case study links the public demo video. It isn't virtual try-on and doesn't promise fit. The case study is about the build, not about users or sales.
