I help organisations put AI into the processes they already run —
choose the right solution and, where needed, build it in.
Copilot · agents · assistants · integrations — wired into real work, measured, handed over.
Services · How I work · Built · FAQ · Book a call
Many organisations have tried AI. Fewer have it in day-to-day work.
- Copilot licences were bought, but usage is sporadic.
- People use AI tools on their own — and nobody knows what data goes in.
- Several offers on the table, no idea which one fits your process.
- The same manual work repeats every day across several systems.
- Leadership expects results from AI, but there's no owner or first use case.
One is enough. I start with one process and take it through to the end. (GitHub won't let you tick these. Email will.)
Tip
Not sure where to start? Describe one process and I'll think it through with you.
A tool alone doesn't change the work. Four things decide whether a pilot stays a trial:
|
If nobody knows which files the tool can see, nobody trusts it. Permissions get fixed before a wider rollout. |
No agreement on which steps a person confirms and which the system takes alone — so it stays a trial. |
|
A licence or a model changes nothing until it's tied to a real task, real data and a named owner. |
No before/after, no basis to decide whether to continue, expand or stop. |
| Principle | In practice |
|---|---|
| 🇪🇺 Data location | Your Microsoft 365 tenant, an EU cloud, or your own server with an open model — written into the proposal. |
| 🔑 Permissions before Copilot | Access is reviewed first, so AI doesn't surface documents people shouldn't see. |
| 🤖 Agents within agreed limits | Defined tools and permissions. Consequential steps are approved by a person. Every action is logged. |
| ⚖️ GDPR & EU AI Act | Considered from the design stage, not bolted on afterwards. |
What "within agreed limits" looks like in practice:
sequenceDiagram
autonumber
actor O as Process owner
participant A as AI agent
participant S as Your systems
participant L as Audit log
O->>A: Task + mandate (tools, permissions)
A->>S: Read within scope
A-->>O: Proposes a consequential step
O->>A: Approve ✅ or reject ❌
A->>S: Act
A->>L: Action logged
I start where you have the most manual work or uncertainty — and take it through to the end.
|
Map processes, data and risk. Leave with a clear recommendation on where to start.
See also: Prompt engineering |
Connect AI, Copilot and agents to the systems you already have — the work stays where it runs.
See also: Copilot rollout · AI agents |
|
Define what the solution must do, write the requirements, compare offers and tools.
|
Can AI search find you, and can agents actually use your site?
|
Note
The EU AI Act review is an initial risk assessment, not legal advice.
| Layer | Hands-on with |
|---|---|
| Assistants | Microsoft 365 Copilot Claude Gemini Grok Manus Meta AI |
| Agents | Custom-built Hermes Agent OpenClaw Copilot Studio |
| Channels | Email Slack Teams Web Chat tools |
| Cloud AI | AWS Azure EU cloud Your own server |
| Models | Open-weight more control, self-hostable · Closed ready-made, different cost & data model |
| In production | Monitoring & logging Agent observability LLMOps |
flowchart LR
M["<b>1 · Map</b><br/>tasks · data · risk"] --> S["<b>2 · Select</b><br/>tools · vendors · 1–2 use cases"]
S --> I["<b>3 · Integrate</b><br/>systems · permissions · owners"]
I --> H["<b>4 · Handover</b><br/>playbook · metrics · owner"]
H --> Y(["Yours. No lock-in."])
classDef step fill:#fdfbf7,stroke:#d95d39,stroke-width:2px,color:#1a2a3a
classDef done fill:#d95d39,stroke:#d95d39,color:#fdfbf7
class M,S,I,H step
class Y done
linkStyle default stroke:#d95d39,stroke-width:2px
Permissions, guidance and owners go in before the circle of users grows. The solution stays with you and doesn't depend on me.
Public products, built and run by me:
|
Search and analysis of public tenders. |
Open connector linking AI assistants to EU tender data. |
Experiments in how an agent works within a human decision and mandate. |
Alvar Laigna — independent AI expert and fractional CTO. 20+ years of software development, machine learning and systems integration. I assess, build and hand the work over so your people can carry on without me.
Pearls and Roses OÜ (est. 2019) grew out of software development, gamified solutions (games included), DevOps and fractional-CTO work. Now: AI that survives contact with Monday morning.
✍️ alvarlaigna.com · Medium · LinkedIn · Agentic Gov
Organisations I've crossed paths with in my work over the years: Swedbank · SEB · Nasdaq · Telia · Elisa · Tallink · ERR · University of Tartu · Starship Technologies · Rovio · Thermory · L'Oréal
📍 Where — Tallinn, Estonia · working in Estonian & English (interactive map)
{
"type": "FeatureCollection",
"features": [
{
"type": "Feature",
"properties": {
"name": "Pearls and Roses OÜ",
"note": "Tallinn, Estonia · ET / EN",
"marker-color": "#d95d39"
},
"geometry": { "type": "Point", "coordinates": [24.7536, 59.4370] }
}
]
}
What does it cost?
Hourly rate €360; a lower day rate applies on longer projects. You get a precise proposal after the first conversation.
How long does it take?
A single-process solution usually takes a few weeks. More systems and more complex logic take longer. The timeline is agreed before work starts.
Do we need our own IT team?
No. You need someone who knows the process and access to the relevant systems. I handle the technical side — and I map, build and hand over myself.
Which systems can AI connect to?
Wherever your work already lives: Microsoft 365, CRM, ERP, accounting, document management, databases, public APIs, AI services on AWS and Azure, plus email, Slack, Teams, the web and chat tools. Before building, I check which integrations are needed — and safe.
Is Copilot worth it?
Not always. Before expanding licences, I test it with a small group and measure whether it saves time on your actual tasks.
Does our data leave the EU?
It doesn't have to. If EU processing is a requirement, I choose a setup that allows it and write into the proposal where data is processed.
I'm not technical. Is that a problem?
No. Your part is to describe what isn't working. I explain the options in plain language, in writing — in Estonian or English.
What about superintelligence?
AI that exceeds humans in most intellectual tasks doesn't exist yet, though progress is fast. Today's value comes from using AI well: tidy data, clear processes, accountability. Conveniently, those are also the preconditions for whatever comes next.
Thirty minutes on where AI would help your work — and an agreed next step.
✍️ You describe the situation → 💬 We talk for 30 minutes → 📄 You get a concrete next step, in writing
…even if it turns out AI isn't the answer. Replies usually within a couple of working days.
Pearls and Roses OÜ · est. 2019 · Tallinn · ET / EN · ↑ top
🌹 Off the clock
Same company, different department: weird little merch for strange times, printed on demand in the EU → store.pearlsroses.com