Periscope

Mon, 01 Jan 0001 00:00:00 +0000

Generate synthetic support tickets for training human or AI support agents.

Overview

Training support has a data problem. The tickets you’d most want someone to practise on are your real ones, and those are full of customer names, order numbers, and details you can’t hand to a new starter or drop into an eval harness. Sanitising them is slow work, and what you’re left with is thin and out of date.

Periscope skips that step. You give it a URL and it reads what your customers already read: your help centre, docs, blog, and public social profiles. Then you answer a short set of questions about the queue you want — which issues come up, which channels they arrive on, how the customer sounds when they’re annoyed. It writes tickets from that, labelled by type, channel, sentiment, and difficulty, one at a time or a few hundred in a batch.

That’s useful in two places. A new rep gets a queue that looks like week three rather than week one, with nothing real on the line. And if you’re shipping an AI agent, you get an evaluation set built from your own product — refunds, escalations, the ambiguous ones — before the agent meets them in production.

No customer conversations are uploaded, and there’s nothing to connect to your live helpdesk. Pricing is credits rather than seats, so a whole team can practise on the same workspace.

Features

Learns from your public content — Point it at a URL and it reads the help centre, docs, blog, and public social profiles
Queues you shape — Choose the issue types, channels, and customer tone you want to practise against
Labelled tickets — Every ticket is tagged by type, channel, sentiment, and difficulty, so you can filter and sort a set
Batch generation — Make one ticket to check the tone, or a few hundred to build a training set
Export anywhere — JSONL, Markdown, or CSV, ready for an eval harness or a training doc
No customer data — Public content only; nothing to sanitise and no helpdesk integration to set up
Credit pricing — Pay for what you generate rather than per seat

Jellybook

Mon, 01 Jan 0001 00:00:00 +0000

Jellybook is an iOS app for Jellycat collectors to catalogue, track, and share their collection. I built it for my fiancé and we run it together — our first iOS app, and close to 1,000 collectors signed up in its first month.

Overview

Jellybook gives collectors a searchable catalogue of thousands of Jellycats, a way to track what they own with notes and photos, a wishlist, and a community “Jellywall” to share their collection. The app itself is native Swift and SwiftUI.

The interesting part is the back end: it runs entirely on Cloudflare. Workers serve the API, D1 holds the data, R2 stores the images, and Workers AI plus Vectorize power image-based search, with Queues handling the background work. There are no servers to manage, it’s fast everywhere, and it costs almost nothing to operate.

Features

Native iOS — Built in Swift and SwiftUI for a fast, native feel
Catalogue — Browse and search thousands of Jellycats by category
Collection tracking — Record what you own with notes, photos, and dates
Wishlist — Keep a list of the ones you’re still hunting for
Jellywall — Share your collection with the community
All on Cloudflare — Workers, D1, R2, Workers AI, Vectorize, and Queues — fast and cheap to run

Projects on Frontier.sh

Periscope

Overview

Features

Jellybook

Overview

Features