Copenhagen · AI adoption, automation and enablement

I help peoplework betterwith AI.

I'm Aðalsteinn, a Business Economics and IT student in Copenhagen and an AI & Automations Student Specialist at Veo Technologies. I interview people to find the work that eats their time, build AI-assisted tools that take it away, and run workshops so they can use AI themselves. Three of those tools are below as case studies, each with a working demo.

  • Discovery
  • Building
  • Enablement
  • Stakeholders
Numbers in. Story out.Warehouse data, manager notes and an AI summary
  • 1 h → 15 min monthly report prep
  • 2 h → 20 min campaign outreach
  • 150 leads from 2 conventions
  • 4–6 h saved per week on meeting follow-up
  • 2 AI agent workshops across 3 regions

Selected projects

Three tools, two agents and a workshop

I built each of these in my role at Veo Technologies. Scroll through the case studies and try the demo beside each tool. The demos use a fictional coffee roaster, Northbrew, with made-up names and numbers. They contain no Veo data, and nothing is sent or saved.

01Case study · Reporting

A monthly report app for every market

It pulls the numbers, the manager adds the story, and AI ties the two together. I built it for the partner marketing team at Veo Technologies.

Problem

Every month, managers in 12 markets put together a partner performance report by hand. They copied figures out of the data warehouse and spreadsheets, listed their activations from Asana and wrote the story around them. It took about an hour each, every report looked different, and the same metric could mean two things in two markets.

Solution

One app where the numbers are already there. The manager opens their market, pulls the month's activations with a click, adds context, and watches the finished report build itself in one shared format. An AI draft connects the big swings in the numbers to what the manager wrote.

What I built

  • A market overview showing which reports are not started, in draft or complete
  • Revenue, ARR, new accounts, leads and top partners pulled automatically, with targets and month-over-month change
  • One-click import of the month's activations from Asana
  • An AI-drafted summary the manager can accept, edit or discard
  • A live preview of the finished report while the manager types
  • "Mark complete", which freezes the figures, with editing rights set per market
  • A page documenting every metric and its source, plus the user guide

Numbers

  • Report prep went from about 1 hour to 15 minutes per manager
  • Used across 12 markets
  • 42 metrics defined and documented in one place

Tech stack

  • Python, FastAPI
  • BigQuery, Firestore, Asana
  • Gemini for the summary
  • Google Cloud Run, Terraform, built with Claude Code

Outcome

  • Live in production at Veo
  • Every market now reports in the same format, with the same definitions
  • Managers spend their time on the story, not on copying numbers

Try it: this is a working, simplified demo with made-up data.Open full screen ↗︎

02Case study · Automation

A campaign outreach cockpit

From a partner list to personal emails with the right links, reviewed by a person before anything is sent.

Problem

For each partner campaign, managers wrote dozens of near-identical emails, every one with that partner's own link, copied across from two spreadsheets. It took about two hours per campaign. Copy-paste mistakes sent partners the wrong link, and nobody could see who had already been contacted.

Solution

A platform that matches the partner list to the campaign's links, drafts one master email with AI and fills in each partner's details. The manager reviews and approves the email, checks every partner's version, and only then does anything get sent.

What I built

  • Campaigns published by an admin, with each manager working in their own region
  • Matching of the master partner list with the campaign sheet, with a plain reason for every partner that is skipped
  • An AI-drafted master email in the partner's language, with one-click revisions, undo and an approval step
  • A preview per partner that can be edited by hand or reset to the master
  • Test mode that sends only to yourself, with a warning before leaving it
  • A check before sending that every email carries its own link and nobody gets it twice
  • Tracking of delivered, opened, bounced and replied, written back to the sheet as "partner contacted"

Numbers

  • Outreach went from about 2 hours to 20 minutes per campaign
  • 5 steps from partner list to tracking
  • Every email is approved by a manager before it is sent

Tech stack

  • Python, FastAPI, JavaScript
  • Google Sheets API
  • Claude for the draft, SendGrid for sending and tracking

Outcome

  • Live in production at Veo
  • Each partner gets their own link, checked before sending
  • The team can see who was contacted and what happened to each email

Try it: this is a working, simplified demo with made-up data.Open full screen ↗︎

03Case study · Events

Spin to Win: a lead game for event stands

Visitors sign up on their phone, spin a wheel and collect a prize at the stand. The team goes home with real contacts.

Problem

At convention stands, leads went onto paper or into a simple form that filled up with fake contacts. There was no fair way to hand out merch, and the popular prizes were gone on day one.

Solution

A registration game. Staff set up the event and the merch they brought. Visitors scan a QR code or use the stand's tablet, sign up, answer a few questions and spin. They get a one-time code, and staff check it before handing anything over. Every sign-up lands in a lead list the team can follow up on.

What I built

  • Event set-up by staff: language, expected sign-ups, merch stock and a deletion date
  • Sign-up on the visitor's own phone or on a kiosk tablet, in English or Polish
  • Contact checks that block duplicates, fake phone numbers and throwaway email addresses
  • A spin decided on the server, which paces the prizes so the stock lasts the whole event
  • Pickup codes that work once and are redeemed by stand staff
  • A live lead dashboard with stock levels and a spreadsheet export
  • A privacy notice and automatic deletion of leads after the set date

Numbers

  • 150 leads from 2 conventions, up from none before
  • 2 languages: English and Polish
  • 100 automated checks run the whole flow end to end before each event

Tech stack

  • Cloudflare Workers, D1 (SQLite)
  • JavaScript, libphonenumber
  • Built with Claude Code

Outcome

  • Used by Veo at 2 conventions
  • Leads go straight into a follow-up list, not onto paper
  • Prizes are only handed out against a valid, unused code

Try it: this is a working, simplified demo with made-up data.Open full screen ↗︎

04Case study · AI agents

Team knowledge agents

Two AI agents that keep a team's knowledge base useful: one answers questions from it, the other keeps it up to date.

Problem

Valuable team knowledge was spread across Slack conversations. Colleagues struggled to find answers, and useful information was rarely captured for the next person who needed it.

Solution

Two AI agents connected to the team's Notion knowledge base. One answers colleagues' questions using the existing documentation. The other turns useful Slack discussions into structured entries, checks for duplicates and adds them to the knowledge base.

What I built

  • An agent that answers colleagues' questions from the documentation the team already has
  • An agent that turns useful Slack discussions into structured knowledge-base entries
  • A duplicate check before anything new is added
  • Controlled write access, so the agents cannot damage existing content

Tools

  • Notion AI custom agents
  • Notion knowledge base, Slack

Outcome

  • A simpler way for the team to find information
  • A knowledge base that stays up to date as the team talks
  • Write access is limited, which protects the content
Two agents, one knowledge base
The answering agentA colleague asks a question. The agent replies using the team's existing documentation.Reads only
↑ readsNotion knowledge basewrites ↑
The capturing agentTakes a useful Slack discussion, writes it up as a structured entry, checks for duplicates, then adds it.Controlled write access
↑ Useful discussions in Slack

No demo for this one: the agents run inside the team's own workspace.

05Case study · Enablement

Notion AI agent workshops

Two hands-on sessions where colleagues at Veo built their own AI agents for the tasks that fill their week.

Problem

Colleagues were curious about AI agents but did not know where to start, and general advice is hard to apply to your own job. The team is also spread across Europe, the US and Asia-Pacific, so a single session would have left people out.

Solution

Two one-hour workshops on the same day, one in the morning and one in the afternoon, so every time zone could join. I explained how agents work, built one live, and then everyone built an agent for a task from their own role while I helped with questions, connections and troubleshooting.

What I did

  • Ran two one-hour sessions for Veo's Markets team on 22 October 2026
  • Built a weekly work-brief agent live, from an empty page
  • Coached people as they built agents for their own partner and market work
  • Wrote a step-by-step walkthrough, from giving an agent context to building a custom one
  • Made the sign-up form, the slide deck and a weekly-brief prompt anyone could copy
  • Recorded the morning session

Numbers

  • 2 workshops, one hour each
  • About a dozen colleagues signed up, across 3 regions
  • 3 agents built or demonstrated by colleagues during the sessions

Tools

  • Notion AI custom agents
  • Notion pages and forms, a slide deck, a shared prompt

Outcome

  • People left with an agent for their own work, not just notes
  • Feedback from colleagues: the team had "definitely benefited", and the guide was "super helpful"
  • The walkthrough and the copyable prompt were shared with the team
Workshop format · 60 minutes
  1. How agents workWhat an agent is, what it can read, and what it can do on its own
  2. Live buildA weekly work-brief agent, built from an empty page
  3. Build your ownEveryone picks a task from their own role; I help with questions and troubleshooting
  4. Show and tellColleagues show what they built

Agents colleagues built

  • Partner-meeting preparationBuilt and tested during the session
  • Daily partner sales updatesStarted during the session
  • Emails and meeting notes into AsanaRuns on a weekly schedule; demonstrated to the group

No demo for this one: it was a live session.

How I work

Find the problem, build the fix, teach the team

01 · Discovery

Start with the people

I interview colleagues and map their repetitive workflows step by step, to find where AI and automation would really help.

02 · Building

Build with AI, check like a sceptic

I build AI-assisted workflows and apps with Claude Code. I review what it writes, test the real flows end to end, and put test modes, access rules and data deletion in from the start.

03 · Enablement

Teach, don't gatekeep

I run workshops on AI agents and write guides and shared prompts, so colleagues can apply AI in their own work without me.

04 · Stakeholders

Keep everyone in the loop

I work with market managers, data analysts, legal, revenue analysts and IT to clarify requirements, improve data definitions and turn feedback into the next version.

About

From fish production to AI tools

I'm a Business Economics and IT student with a strong interest in using AI to solve practical problems and help people work better. As an AI & Automations Student Specialist at Veo Technologies, I interview colleagues about their challenges, build AI-assisted tools and lead workshops so people can explore AI in their own work.

Before moving into AI, I spent seven years in fish production for Iceland's largest fisheries company, at times as production lead, and as ship's cook with my own budget, procurement and supply logistics. It made me comfortable in demanding, fast-moving environments, and it is why I care that a tool holds up in real work.

Beyond the three demos, I automated meeting follow-up: emails to stakeholders drafted from AI meeting notes, with decisions and action items in one consistent format. It saves 4–6 hours a week.

  • Claude Code
  • ChatGPT
  • Gemini
  • Cursor
  • n8n
  • Zapier
  • Power Apps
  • BigQuery
  • Power BI
  • Google Cloud
  • Salesforce
  • Asana
  • Notion
  • GitHub
Now
AI & Automations Student Specialist, Partner MarketingVeo Technologies · Aug 2026 – present · Copenhagen
Studying
Bachelor of Business Economics and Information TechnologyEK – Erhvervsakademi København · 5th semester
Before
Fisherman and ship's cookBrim hf., Iceland · 2017 – 2024
Shift managerDomino's Pizza, Iceland · 2016 – 2017
Languages
Icelandic, English, Danishnative · fluent · professional

Got a workflow that eats hours?

I'd like to hear about it. I'm open to roles in AI adoption, automation and enablement.

Email me ↗︎