Prepared for the AI & Automation Specialist role

A planning companion your couples would appreciate, and the organising your team will enjoy

An assistant system for The Gilchrist Collection. It handles routine questions, sheets and follow-ups so coordinators spend their time with couples, and a research lab tests new ideas for the business. Shown as three working sketches by Linards Berzins.

See the product What I’ve built
My AI assistant for research and organising
In daily use since April 2026

Checks my email, calendar, Slack and CRM every 2 hours. Over 2,000 automated checks run against it.

Average UK wedding, 2026
£20,604

Bridebook, about 7,000 couples

Couples who trust AI to help find a venue
1 in 5

Bridebook UK Wedding Report, April 2026

Weddings a year across the group
2,500+

The Gilchrist Collection, CEO page

01Product line

A product line, from one set of venue knowledge.

Eleven ideas, grouped by who pays. They share one asset a generic app lacks: how a venue group actually runs, which is the first thing I would learn from you. The three marked below are the ones I sketch in the next section.

For couples

Sold as package extras or included to win the booking.

  • Ask your venue

    A planning companion that answers from the venue guide and the couple’s own booking, and hands the rest to their coordinator.Sketch 1, built next

    Why Gilchrist: Answers come from your own guides and contracts, not the open internet.

    Weeks

  • See your room in your colours

    Pick palette, flowers and layout; the room re-dresses with an indicative price, and on site it shows through the phone camera.Sketch 2, built next

    Why Gilchrist: Only items your venues and suppliers can actually provide.

    A few months

  • Your day, timed for your venue

    A timeline built from real room turnarounds, kitchen times and sunset, with clashes explained in plain words.Sketch 3, built next

    Why Gilchrist: If you hold real timings from past weddings, a generic app cannot match them.

    A few months

  • Guest concierge on the day

    Guests ask about parking, food, taxis and dress code and get answers from approved information only.

    Why Gilchrist: Guests get quick answers on the day, so the couple and the team can enjoy it.

    Weeks

  • Memory book after the day

    Photos, speeches and the day’s timeline gathered into one keepsake, with no face matching at launch.

    Why Gilchrist: You are there at the moment couples care most.

    A few months

For your venues

Tools that help your teams, with people always in charge.

  • Enquiry and viewing assistant

    An instant, accurate reply to every enquiry, with real availability and viewing slots.

    Why Gilchrist: It would use your own prices, availability and the replies that have worked, if you keep them.

    Weeks

  • Enquiry to booking, one view

    One live view of enquiries, tours and bookings for each venue, plus faster first replies.

    Why Gilchrist: From your public sites, each venue appears to have its own brochure form and tour booking, so joining them up is a group job.

    Weeks

  • Run sheet drafted from the couple’s choices

    The staff briefing built from the couple’s own choices, with missing items flagged a week out.

    Why Gilchrist: It would draft the sheets your teams already write by hand, from the couple’s choices, for them to check, if they do.

    Weeks

  • Seating and dietary sheet

    Diets, allergens and table plan arrive at the kitchen as one sheet, signed off by the head chef.

    Why Gilchrist: Built around each venue’s rooms and kitchens, with a person always signing off.

    A few months

For the business: an innovation lab

Helps Gilchrist spot what couples want next, and test it before building.

  • Wedding innovation lab

    A research loop that follows wedding trends, what couples ask for, new package ideas and competitor moves, and tests each idea before anyone builds it.

    Why Gilchrist: Twelve venues and 2,500+ weddings a year are a rich source of questions, if you want them studied. Set up in October 2026, first questions running.

    Weeks

Later, for other venues

Only after it works across your own twelve.

  • The same kit for independent venues

    The enquiry, timeline and concierge tools packaged for venues that run on their own.

    Why Gilchrist: A playbook that would first be tested on your own sites.

    A few months

Size is my first estimate: weeks, or a few months. It is a sketch, not a quote. Which data you hold today is the first thing I would check.

02The product

Companion: one product, two sides.

Couples get a planning companion from the day they book. Your teams get clearer requests and more time with couples. Everything is a sketch for a made-up demo venue, with pre-written answers and invented rooms and prices.

What couples get

  • Answers in plain words, each one showing where it came from.
  • A visual plan: their room in their colours, and a day timed for the venue.

What your teams get

  • One daily summary of every question, showing what was answered and what needs a person.
  • One quote request, written up from the couple’s choices, for a coordinator to check.
  • Run sheets, supplier sheets and kitchen sheets drafted from the couple’s choices, with fewer slips. People check every one.

A small impact model

Change any number to see the effect. The 2,500+ weddings a year is from your website; every other number is an example, so swap in your own.

From your website (2,500+ a year). Change it if you know better.
Example. Use your own number.
Example. Use your own number.
Example. Use your own number.
Example. Use your own number.
Example. Use your own number.

Not a forecast, and it ignores build and running costs.

Example: hours freed for couples a year1,667

Example: add-on revenue a year£37,500

Example: time on routine replies today4,167

Time freed for couples1,667

Hours freed = weddings × questions × minutes × share answered straight away ÷ 60. Revenue = weddings × take-up × price.

Sources

Ask your venue

Sketch: a demo venue with invented rooms and made-up prices. Not a Gilchrist venue, guide or quote.

The problem A couple wants a straight answer about their own day, without waiting for an email reply.

What I’d test first Do couples trust an answer more when it shows the page it came from, and how many questions still need a person? Not tested yet.

This sketch needs JavaScript to be tappable.

The couple’s phone

Screen 1 of 5

Ask about your day

Amy and Sam. Saturday 14 November, The Glasshouse at Fernhill (demo venue).

Try asking

Sketch: it recognises a few words from the demo guide. Anything else goes to Hannah, your coordinator.

What lands on the team’s desk

Hannah’s daily summary for Fernhill (demo), Monday 5 October. It updates as the couple taps. One summary of every question, so coordinators see it all at a glance.

Answered automatically: 0
    Needs you: 0
      Deadlines
        The route through it You’re on: Home
        1. Home every screen leads back here
          • If the guide covers it: Answer with source
          • Costs money: Add 10 evening guests
          • Deadline reminder: Final numbers
          • Not in the guide: Passed to Hannah
        2. Answer with source
          • Not what I meant: Passed to Hannah
        3. Costs money indicative price
          • Ask for OK: Passed to Hannah
        4. Deadline reminder
          • Add guests: Costs money
        5. Passed to Hannah written up, reply time given

        My choices behind this sketch

        1. Answers only from our own documents

          What I chose
          Answers come from the venue guide and the couple’s own booking, never the open web.
          What I ruled out
          A general chatbot that will answer anything, including things the venue never said.
          How I’d count it
          Share of answers a coordinator marks correct in a weekly sample.
        More of my choices

        1. Every answer shows where it came from

          What I chose
          Each answer carries a small label showing the document and page.
          What I ruled out
          Plain answer text that the couple has to take on trust.
          How I’d count it
          How often couples still email to double-check an answer that has a source.

        2. A person is one tap away

          What I chose
          What is not covered goes to the named coordinator, written up, with a reply time.
          What I ruled out
          A loop of “I did not understand” with no way out.
          How I’d count it
          Questions per wedding that reach a coordinator, and minutes each takes.

        3. Anything that costs money needs a yes

          What I chose
          The price is shown as indicative and the change waits for the coordinator’s OK.
          What I ruled out
          The assistant confirming extras by itself.
          How I’d count it
          Approvals answered within one working day, and extras sold per wedding.

        See your room

        Sketch: an invented demo venue with made-up prices and dates. Not a Gilchrist venue, price list or quote.

        The problem Picture our wedding in a room we can actually book, and know roughly what it will cost, before we visit.

        What I’d test first Do couples who see the price build as they choose send fewer changes after the viewing? Not tested yet.

        Step 1 of 5Couple’s phone
        Illustration: the room as you have dressed it.

        Pick a room and a date

        Three rooms at the demo venue

        Sketch: in the venue, the same view would run through your phone camera. Not built.

        On the team’s desk

        Preview, not sent yet

        Nothing reaches the team until the couple taps Send to my coordinator.

        The route through it You’re on: Room and date
        1. Room and date three rooms, three dates
        2. Palette four named palettes
        3. Flowers and candles three styles
        4. Tables and guests
          • Fits: Indicative price
          • Too many guests: Over capacity
        5. Over capacity a plain suggestion
          • Take the suggestion: Indicative price
          • Change the guest count
          • Pick another room myself
        6. Indicative price item not available: swap offered
          • Item swapped: price updates
          • Send to my coordinator
        7. Sent to coordinator
          • Start again: Room and date

        My choices behind this sketch

        Whole product

        What I chose
        Only show what the venue can deliver: real rooms, real linen, real florist stems.
        Why
        A couple who falls for something we cannot supply is a worse conversation than none.
        What I ruled out
        Generic AI images of any wedding the couple describes.
        How I’d count it
        Share of requests the coordinator confirms without changing an item.
        More of my choices

        Screens 2 to 5

        What I chose
        The price is visible from the first choice and builds line by line.
        Why
        Couples decide differently once they see what each choice costs.
        What I ruled out
        A request-a-quote form with the price arriving days later.
        How I’d count it
        Quotes revised after the first reply, before and after.

        Screens 5 and 7

        What I chose
        The picture becomes the quote: one request with room, choices, items and total.
        Why
        The coordinator stops re-typing what the couple already showed.
        What I ruled out
        The couple describing their picture in an email.
        How I’d count it
        Coordinator minutes from enquiry to a confirmed quote.

        Screen 6

        What I chose
        Over capacity says what would work, such as long tables or another room.
        Why
        A bare error ends the session; a suggestion keeps it going.
        What I ruled out
        A “too many guests” error with no next step.
        How I’d count it
        Share who take a suggestion rather than leave.

        Your day

        Sketch: a demo venue with invented rooms, rules and times. Not a real Gilchrist venue.

        The problem Know the day will work before it is booked, and give the team accurate sheets drafted from the couple’s choices.

        What I’d test first Do couples fix a flagged clash with one tap rather than writing to the coordinator? Not tested yet.

        Screen 1 of 3

        Your choices

        Starting points are filled in. Change anything and see the day rebuild.

        Where is the ceremony?

        On the team’s desk

        Three sheets written from the couple’s choices. They update as the couple changes them.

        Run sheet
          Supplier arrivals
            Kitchen
              See the full sheets
              The demo venue’s rules
              • Ceremony lasts 30 minutes. The Glasshouse needs 60 minutes to turn from ceremony to breakfast.
              • Photos start 15 minutes after the ceremony and take 30 minutes. They should finish at least an hour before sunset.
              • Kitchen needs 2 hours 15 for three courses for up to 120 guests, 2 hours 45 above that. Breakfast and speeches take 2 hours 30.
              • Evening guests need at least 3 hours before the bar closes at midnight. Carriages at 12:30am.
              • Sunset is a typical mid-month time for central England, from a small table.
              The route through it You’re on: Your choices
              1. Your choices ceremony time, month, room, breakfast, guests
                • See my day: Your day
              2. Your day clashes flagged in plain words; the team’s sheets update beside it
                • One-tap fix: the day rebuilds
                • Late change: Late change, 7 days out
                • Change choices: Your choices
              3. Late change, 7 days out what updates, who is told
                • Back to your day

              My choices behind this sketch

              Built from the venue’s real timings

              What I chose
              The timeline is computed from the venue’s own rules: room turnaround, kitchen time, sunset and bar close.
              What I ruled out
              A generic wedding-day template that looks right and is wrong for the Glasshouse.
              How I’d count it
              Timeline changes made by the coordinator in the last fortnight before the wedding.
              More of my choices

              One set of choices, three sheets

              What I chose
              Run sheet, supplier arrivals and kitchen sheet all regenerate from the couple’s choices, so one change reaches everyone.
              What I ruled out
              Three documents typed up by hand and kept in step by email.
              How I’d count it
              Staff hours per wedding spent on run sheets and re-issuing them.

              Allergens always need a person’s sign-off

              What I chose
              The system lists and counts allergens. It never clears them: each is marked as needing the head chef’s sign-off.
              What I ruled out
              Letting the system swap dishes or mark an allergy as handled.
              How I’d count it
              Allergens signed off before the final numbers date. The target is every one.

              A clash comes with a fix

              What I chose
              Each clash is said in plain words with one tap to fix it. If there is no fix, it says so and points to the coordinator.
              What I ruled out
              A red warning with no way out, which becomes an email.
              How I’d count it
              Share of clashes the couple fixes without writing to the coordinator.
              03How I work

              How I work: start with the work, not the tool.

              I begin by sitting with the people doing the job and watching how it really goes. Then I count where the time goes, pick the jobs that pay back most for least effort, and build the smallest useful version. The people who use it test it, and I measure the result: time saved, errors, revenue. If it can be counted, it gets a number.

              Simplest answer first. Sometimes a better form or a change to the process beats AI, and then that is what I recommend.

              1. WatchSit with the team and see the workMe, on site
              2. MapWhere the time goes, step by stepMe, with the team
              3. PickChoose by impact and effortAgreed with you
              4. BuildThe smallest useful versionMe
              5. TestWith the people who use itThe team
              6. MeasureTime saved, errors, revenueCounted, reported
              7. RunKeep it working after launchMe
              8. ImproveWhat the numbers show goes back inNext pass
              The rule for any AI I build: it drafts, a person decides, and it never touches customer data it does not need.
              04What I've built

              What I've built.

              Every system below I built myself, and each one follows the same rule: the AI drafts and a person decides. None has a result measured in hours yet, so each one says what you can actually see, not a made-up number.

              The lead example

              Fredis, my AI assistant for research and organising

              • Problem: checking email, calendar, Slack and customer records, and researching people before meetings, took time every day.
              • What I built: every two hours it reads my email, calendar, files, Slack and CRM (customer records), and flags only what changed.
              • Ask it about a person or a company and it researches them and files the briefing where I can find it later.
              • It drafts replies and follow-ups. I approve every one. It cannot send anything on its own.
              • My part: designed and built by me, using AI coding tools.
              • Result: in daily use since April 2026, running on its own server. What you can see is months of daily use, a growing library of drafts and briefings, and a review queue I work through each day.
              • Over 2,000 automated checks run against it, so a change that breaks something is caught before I rely on it.
              How it fits together
              GmailCalendarDriveSheetsSlackHubSpot
              My AI assistantreads, researches, drafts
              Drafts for my reviewBriefingsAlerts on what changed
              Nothing goes out until a person approves it.

              What it would do here: sort new enquiries across the venues, draft follow-ups for coordinators to check and send, nudge suppliers who have gone quiet, and draft a weekly report for head office, so people have more time for couples.

              Built with: Python, Claude, Google Workspace, Slack, HubSpot

              Innovation

              Research lab for new ideas
              Problem
              New ideas often get built before anyone has checked that people want them.
              What I built
              A research loop. Each idea starts as a question. I research many sources, shape an idea, build a quick prototype, and re-run the research every week. At most three questions are live at a time, so each gets proper attention, and what I learn is kept.
              What it would do here
              Follow wedding trends, what couples ask for, new package ideas and competitor moves, and test each idea before anyone builds it.
              My part
              Set up by me. My AI assistant does the research and I decide what matters.
              Sales follow-up

              Research to follow-up deck
              Problem
              A tailored follow-up after a meeting took hours.
              What I built
              I type my meeting notes into Slack. It researches the prospect and builds a short, tailored deck, shared by private link only after I confirm.
              My part
              Designed and built by me, using AI coding tools. It runs on request.
              Result
              Proven end to end on a demo deck. Not yet timed. A venue tour could work the same way: notes in, personal proposal draft out.

              Built with: Claude, Slack, Google Calendar, Cloudflare Pages

              Customer support

              Shop support assistant
              Problem
              A small online shop needs to answer customers at any hour, when nobody is free to reply.
              What I built
              An online shop, live and taking card payments, with an assistant that answers customers day and night, looks things up, and hands over to a person when it should.
              My part
              Built using AI coding tools.
              Result
              Live and taking payments. I hold no order or revenue figures to quote. It is the same shape as an assistant that answers the first enquiry.

              Built with: Next.js, FastAPI, Stripe, Claude

              Small firms

              Prototypes to show the fix first
              Problem
              Small firms lose time to enquiries, quotes and spreadsheets, but cannot picture the fix.
              What I built
              A website estimator for a roofing firm that captures the enquiry. A clickable tool for a timber trader that imports and exports Excel sheets. A recruitment tool where every claim in a candidate pack is sourced.
              My part
              Built using AI coding tools, from the first questions to the live demo.
              Result
              Unpaid prospect work, so nothing measured. The point is to put something real in front of the people who would use it.

              Built with: HTML, CSS, JavaScript, Python

              Operations

              Bus operations platform
              Problem
              A bus operator needs timetables, vehicles, incidents and punctuality in one place, for different kinds of staff.
              What I built
              Dashboards, roles that control who sees what, two languages, and an AI assistant that advises while a named person decides.
              My part
              Designed and built by me, using AI coding tools.
              Result
              Presented at Riga City Hall in May 2026. Not yet in service. Rotas, incidents and access rules are problems a venue group knows well.

              Built with: Python, FastAPI, PostgreSQL

              Email

              An email production platform
              In one line
              Over 150 ready-made email sections, and AI checks against ten quality tests on every email before anything goes out. A feature that turns Figma designs into email code is in testing. Built using AI coding tools, not yet live with a customer.

              Built with: Python, Claude

              05For this role

              For this role.

              What the advert asks for, and where I have already done it. Where I have not, I say so.

              Find manual, repetitive work and build the fix

              My AI assistant now does my daily trawl through inbox, calendar, files and CRM, and I review what it finds. The prototypes for small firms start from one slow process.

              Workflows, integrations and AI agents

              My AI assistant connects Gmail, Calendar, Drive, Sheets, Slack and HubSpot, and runs on a schedule. In agency work I built steps in automated customer journeys, alongside the journey builders, and the personalised emails they sent.

              Python and JavaScript, databases and APIs

              Every system on this page is built in Python or JavaScript, with databases and outside services connected through their APIs.

              Choosing the right AI for the problem

              I build with Claude and pick the approach by the job. The transit platform and the shop assistant each use AI in a different way.

              Build a rough version fast, test it with the people who use it, then make it dependable

              Prototypes for small firms, shown live before anyone pays. My AI assistant has run daily since April 2026, and over 2,000 automated checks run against it.

              Standards for AI use, data and access

              Every AI I build drafts and a person decides. It cannot send anything on its own, keeps passwords out of what it writes, and has rules on what it may touch.

              Teach colleagues

              In agency work I trained and onboarded junior colleagues and wrote the process documentation.

              Simplest right answer

              Sometimes a better form or a change to the process beats AI. I say so first.

              Find opportunities nobody has listed

              Each venue seems to run its own website, brochure form and booking calendar, so I would start by joining enquiries, tours and bookings into one view.

              Honest gaps

              • I have not used n8n, Make or Power Automate. I have built the same steps in code (start on an event, take different routes, try again if it fails, wait for a person to approve), and picking up a new tool should not take long.
              • I have not measured hours saved on my own tools, so I would take a baseline at a venue first.
              • This would be my first in-house AI role.

              First 90 days

              1. Weeks 1 to 4. Sit with sales, coordinators and finance at head office, and with two venues. Map the journey from enquiry to wedding day, and the time each step takes.
              2. Weeks 5 to 8. Ship one quick win, measured. Most likely a single enquiry-to-tour-to-booking view across the venue sites, with faster enquiry replies and follow-ups.
              3. Weeks 9 to 12. Prototype the first piece of Companion with one venue.

              Local: I live in East Grinstead.

              Contact: linardsberzins@gmail.com · 07949 783 799 · CV (PDF)