AI Tech Rave 2.0 · Warsaw

Oscar AI

The honest story of an internal AI Challenge — from day zero to production

Who I am

PM at Payment Infra Stream — Cards · APMs · Reconciliation.
One of the organizers and judges of the internal AI Challenge (Oscar).
5 years at Solidgate.
Fun fact: ran 500+ km last December.

Why we started this?

00 / Origin

What was our goal?

The idea started as a test internal initiative under the working title "Oscar for AI projects" — an AI-project competition in an awards-ceremony format.

Goal 1

Give people a better sense that AI isn't scary.
Even if you've never worked with it — you can try and get it wrong.

Goal 2

Popularize AI adoption through a competitive element.
Limited-edition prizes — a reason to finally give it a try.

It started as just an idea — no budget, no deadline, no guarantee anyone would show up. The main goal: prove that AI is "not scary".

01 / Prep

Before competing — training

Before the challenge kicked off, we ran dedicated workshops for teams — to help colleagues find exactly which tasks in their own work AI could automate or improve.

workshops by function: Sales, BizDev, LAC, Marketing + Tech writing, engineering teams, Compliance + Risk

100%

of workshop recordings landed in the shared Learning Hub

~3 weeks

from the first workshop to registration opening

The point of the workshops wasn't to teach "prompting" — it was to find, together with the team, one concrete task from their own daily routine, to show that AI speeds things up right here, right now.

02 / Challenge

How the challenge played out

teams registered to take part

nominations:

Merchant or Product Impact
Internal Efficiency
Bold Bet

final night — live pitches and winners announced in every nomination

The final format was a spoken pitch, no slides on screen: a team comes up and in a few minutes shows what they built.

02 / Challenge

"It works. Clear business value. Measurable impact."

That's how the jury framed the winning criterion — not about a pretty idea, but about what actually works.

Who decided

A jury — evaluated all nominations together, against the criterion above. One winner per category.

We shaped the judging criteria so this wouldn't turn into a "best presentation" contest — but a real evaluation of a genuinely solid technical solution. The real bet is whether the project survives after the AI Challenge.

Internal Efficiency

Routing Copilot: routine decisions — no longer on a human

Routing Copilot

Team: Routing

An AI copilot for the Routing team: helps with routing configuration, drafts AI summaries of config changes for the audit log, watches provider caps, and warns about outages. The goal — cut manual work and remove human error.

METRIC: ~300 hours/week saved

What's cool here isn't automation for its own sake — it's that the copilot flags a problem before it becomes an incident, telling you on its own when a cap or config needs attention.

Merchant or Product Impact

AI Payment Analyst for merchants: insights in a single chat

AI Payment Analyst for merchants

Author: Andriy Kukharuk

An AI chat that helps a merchant find any information about their payment analytics in one place — surfacing insights and improving their payment flows.

METRIC: >1500 requests/week through the chat

A merchant no longer needs to dig through dashboards or write to support — just ask the chat, and AI finds the answer in their own payment analytics.

Bold Bet

Contract Copilot: a Bold Bet that actually paid off

Contract Copilot

Author: Lidiia Rachynska

An AI copilot that drafts the contract from the offer on its own: it continuously scans the CSM board and generates a contract draft the moment a Jira ticket gets the #ready-for-generation label. That's exactly why it's a "Bold Bet" — not an engineering team, not a typical AI-friendly area, but a function where trust in automation is traditionally lowest.

METRIC: 30 deals/week processed this way

Bold Bet isn't about technical complexity — it's about the courage to try AI where it's not obvious. And the bet paid off, literally.

03 / Conclusion

From day zero to production, four made it

workshops for teams — took away the fear of AI

teams submitted projects — with their own real tasks

winners — one per nomination

solutions confirmed alive in production:

Routing Copilot for ops teams
AI analyst in Solidgate Hub
Contract Copilot in Jira
Ward × Veritas on-call

The challenge's main result isn't 4 projects in production — it's that 23 teams went from "just heard about AI" to "tried AI on their own real task". And four went a step further — they shipped their solution to production and gave the company real, measurable value.