The Lab

experiments + workshops

We build cool things with AI, then teach how they work.

The lab is the messy end of Estric AI: voice that answers in two seconds, a front desk that writes its own knowledge updates, ads that render themselves. Some of it ships. All of it gets taught, to business owners in one workshop and to kids in another.

Sydney, small groups, and we travel. Dates go to the waitlist first.

The Estric AI Lab: we build cool things with AI, then teach how they work

On the bench right now

Labelled honestly. Live means customers use it today. In testing means we use it on ourselves first. Cooking means it does not work yet.

Live

A phone answered in two seconds

A websocket held open for the whole call so speech goes straight to the model and back. We chase the latency budget one hop at a time, because a pause that a human would not make is the tell.

Live

A front desk that trains itself

Every night it reads the conversations it had, notices what it could not answer, and writes the missing facts up as suggestions. A human approves them. It never teaches itself unsupervised.

Live

A router that picks the cheapest model that still gets it right

Simple questions go to a small fast model, hard ones escalate. Provider-pluggable on purpose, so no single vendor gets to be the whole product.

Live

Language detected mid-sentence

Someone switches to Arabic or Thai halfway through a message and the reply switches with them, on the phone as well as in chat.

Live

A booking that holds itself until the deposit lands

The slot is held, the payment link goes out, and the appointment only confirms when the money arrives. Nobody blocks their calendar on a promise.

Live

A nightly scan looking for trouble

Unusual usage, disputes, patterns that look like abuse. It raises a flag for a person to read. It never suspends an account on its own.

In testing

Ads that generate themselves end to end

Script, voice, avatar, captions, render. Our own marketing is the test bench, so the rule is simple: the founder says it, never a synthetic customer.

In testing

Client pages on their own domain

A booking page that lives at the business’s own address instead of ours, issued and renewed without anybody touching DNS by hand.

Cooking

A notetaker that feeds the front desk

It joins the meeting, writes it up, and turns what was said into answers the customer-facing AI can give once you approve them. Not switched on yet.

The write-ups live on the blog, and the front desk itself is on features.

Two workshops

Same idea both times: make something real in the room, then work out what the machine is actually doing. One for grown-ups, one for kids.

For grown-ups

Build a website with AI

Walk in with an idea, walk out with a live site. Then come back and make it fast, findable, and able to answer customers while you sleep.

Owners, freelancers, anyone with an ideaTwo sessions
  • What an LLM actually is, in plain language, and the two consequences that explain every weird thing it does.
  • The six line brief. Vague in, generic out; this is the fix.
  • Ship the skeleton, then iterate in small specific changes instead of arguing with the machine.
  • Day two: forms, bookings and money without a backend, page speed, and being found by AI search.

Ages 7-13

Kid AI Lab

Kids make things with AI, then learn exactly where it lies. Half the lab is spent breaking it on purpose.

Two age streams, 7 to 9 and 10 to 13Two sessions
  • Build a game from one idea and put it on a link they can send to a grandparent.
  • Train a machine to see, then fool it and work out why it broke.
  • Catch it inventing a festival that never happened.
  • Design an AI helper for a real person, that has to ask a human before anything important.

Inside Kid AI Lab

What they walk out with

Four things, all of them real, all of them theirs. Nothing here is a worksheet.

A game that actually works

Built in the room from one idea, iterated by talking to an AI, live on a link that night.

A machine they trained

It learns their hand, their toy, their drawing. Then they break it on purpose and work out why.

A lie detector for AI

They watch it answer confidently about something that never existed, and leave with four rules for checking.

An AI they designed

One helper, one boring job, one rule: ask a human before anything that matters.

Good robot, bad robot: the lesson deck

Twenty-one slides on what AI is brilliant at and where it lies, with the facilitator script in the presenter notes. Free to run at your school or library.

Open the deck

Two sessions, two different jobs

Making comes first. Judging comes second, once they have something of their own at stake. In that order it lands; the other way round it is a lecture.

Session one

Make it

90 minutes

  • 10 minRace the robot. Kid versus AI, same task, timer running. They win the rounds that need taste.
  • 25 minMonster game. Same prompt twice, vague then specific, and the difference is theirs to spot. Card printed to take home.
  • 35 minOne-prompt arcade. Their game, from an idea, on a live link.
  • 10 minHallucination hunt. Ask about something that does not exist and watch it answer anyway.
  • 10 minShow and tell. Links sent home.

Session two

Teach it

90 minutes

  • 15 minFix the one thing they hated about their game. Small, specific changes.
  • 30 minTeach the machine. Train an image classifier in the browser, then feed it something it has never seen.
  • 15 minSpot it in the wild. Where AI already sits in their day, listed by them, not by us.
  • 30 minDesign the job. Their own AI helper, personality written in one paragraph.
  • 10 minRules card, arcade page, links home.

Four ideas, arriving as games

Every one of these is something they do with their hands, then put a name to. Kids remember what they proved themselves.

The upside

Fast, tireless, willing to try anything

Good for first drafts, boring jobs and getting unstuck. Proved by racing it, and by noticing which rounds a human still wins.

The downside

It guesses, and it never says "I don't know"

Three demonstrations, not three slides: an invented festival, a classifier that is confidently wrong, and a look at who it assumes a doctor is.

The present

They used AI a hundred times this week

Recommended videos, filters, voice assistants, maps, autocorrect, self-checkout cameras. Nobody told them. Now they can name it.

The future

The deciding parts stay human

The boring parts get automated, being good at asking becomes the skill, and someone still has to answer for what the machine does.

The card that goes on the fridge

  1. 1It guesses. Check anything that matters.
  2. 2It never says "I don't know".
  3. 3Never tell it your real name, address, school or password.
  4. 4If a person made it, a person is responsible for it.

Two age streams

A 7 year old and a 13 year old need different rooms. Same ideas, different hands on the keyboard.

Ages 7-9

60 minutes per session

A grown-up is welcome to stay. Prompts are said out loud and typed by the facilitator, so reading and typing speed never gets in the way. More making, more printing, more moving around.

Ages 10-13

90 minutes per session

Hands on the keyboard, running the loop themselves: describe, look, change one thing, look again. Ends with a real debate, usually whether a bus should be allowed to drive itself.

How we run it

The part parents ask about first

AI tools are not built for children. That is a design constraint on the lab, not something to work around quietly.

No child accounts, ever

Almost every AI tool sets a minimum age of 13 or 18 in its terms. Kids work on a lab device signed in to a facilitator account. Nobody signs up, nobody hands over an email.

No faces, no personal details

We never upload a child’s photo, name, school or address to any AI model. Creatures, drawings, hands and objects only.

Written consent up front

One short form covers photography, the take-home links and what gets kept. Anything a child makes belongs to them, and lab accounts are deleted after the second session.

A small room, watched closely

Small groups with an adult facilitator on the floor the whole time, on a filtered network, with only the tools on the lesson plan reachable.

Posters

Hosting a lab at a school, library, co-working space or community centre? Print these. Ask us and we will send the full resolution files with your dates on them.

Questions

What is the Estric AI lab?

Two things under one roof. It is where we build the parts of our AI front desk that are not obvious yet, from voice latency to a system that suggests its own knowledge updates, and it is where we teach that work as short hands-on workshops for adults and for kids aged 7 to 13.

Do I need to be a customer to join a workshop?

No. The workshops stand on their own and nothing is a sales pitch with a worksheet attached. If you run a service business you will recognise the problems, and the tools we use are ones you can keep using afterwards for free.

Do I need to know how to code?

No, for either program. There is no coding in the kids lab, and the adult workshop is built around describing what you want clearly and then checking what came back, which is the part that actually transfers.

What ages is Kid AI Lab for?

Ages 7 to 13, run as two separate streams: 7 to 9 in 60 minute sessions with prompts said out loud, and 10 to 13 in 90 minute sessions with the kids driving the keyboard. Mixing the full range in one room does not work, so we do not.

Do the kids need their own AI accounts?

No, and they will not be given one. Most AI tools require users to be 13 or 18 and over. Children work on a lab device signed in to a facilitator account, so no child ever creates an account or hands over personal details.

Is it safe? What about the data?

We never upload a child’s photo, name, school or address to an AI model. Sessions run on a filtered network with only the tools on the lesson plan reachable, an adult facilitator is present throughout, a written consent form covers photography and take-home links, and lab accounts are deleted after the second session.

Where does it run and what does it cost?

Sydney, in small groups, with dates and pricing announced to the waitlist first. Get in touch and we will send the next dates, the venue and the price before booking opens.

Can you run the lab at our school or workplace?

Yes. Both programs travel, and we will put your dates and your logo on the posters. Schools, libraries, community groups and team offsites all work.

The next lab fills from the waitlist

Tell us which one you want, a grown-up session or a kids one, and roughly where you are. We will send the next dates, the venue and the price before booking opens, and nothing else.

Schools, libraries, community groups and team offsites: both programs travel.