Algirdas Rumšas · Remote-first
Your data should tell you what to do next. I build the layer that makes that possible.
I work with growing companies that are making decisions on gut feeling - or on numbers nobody fully trusts. I build the data infrastructure, metrics, and AI layer that changes that.
Available for full-time roles or hands-on engagements from ~2 weeks/month.
Past and current clients include OrderYOYO, Whatagraph, Cobiro and others.
Selected work
Four examples of what it looks like when data goes from unreliable - or nonexistent - to something the business actually runs on.
- Consulting · Microsoft Fabric
OrderYOYO (return engagement)
- Problem
- Someone had figured out how to game the platform: place an order, dispute the charge, keep the goods. £30k a month was walking out the door, and nobody had a clear picture of who was doing it or how to stop them.
- Approach
- Built a full detection pipeline on Microsoft Fabric to surface the pattern, identify bad actors early, and put a dispute response process in place before the chargeback window closed.
- Result
- Monthly chargebacks dropped from £30k to £5k. The system paid for itself in the first month.
- Past role · 2020-2024
Whatagraph
- Problem
- The business was growing but leadership couldn't trust the numbers. MRR, retention, and activation were calculated differently depending on who you asked, and sales had no way to know which customers were ready to buy more.
- Approach
- Built the metric layer from scratch - one definition, one source of truth. Then built models to score which customers were likely to upgrade or churn, so sales and CS could act on it. One of those models shipped as a product feature.
- Result
- Monitoring leadership could rely on. ML that started as an internal tool and became a customer-facing product.
- Product
Forbi
- Problem
- Power BI teams spend years building a trusted metric layer - then employees ignore it and ask AI to do the maths instead. The AI guesses. The answers drift from the approved numbers. Nobody catches it until it matters.
- Approach
- An AI assistant that works with your existing Power BI semantic model, not around it. Answers come from the metrics your team already defined and approved - so the numbers are always consistent, not generated on the fly.
- Result
- Self-serve analytics that actually uses the governed layer, with AI risk contained to 'which metric?' not 'is this number right?'
- Product
ntxt
- Problem
- Every AI session starts from zero. You re-explain your context, your constraints, your decisions - and the moment you close the tab, it's gone again. Across multiple tools, this compounds fast.
- Approach
- A persistent memory layer that connects to your AI tools. It builds a graph of what you've decided, what you're working on, and what matters - and makes it available wherever you're working.
- Result
- Your AI tools remember what they need to know. You stop re-explaining yourself.
Get in touch if any of this sounds familiar - or if you want to see the fuller picture.
About
I started out as a Financial Assistant at OrderYOYO. When they decided to build out BI, they asked me to own it - despite that not being my job. I said yes, built the whole thing from scratch, and haven't looked back. That pattern has repeated at every company since: arrive, find chaos, build something that works.
What I care about is data that actually changes decisions - not dashboards that get opened once and forgotten. I get frustrated when analytics is treated as a reporting service. At its best, it finds things leadership didn't know to ask, gives sales a list of who to call today, or turns an internal model into something customers pay for.
Contact
If you need a senior IC who can span lakehouse, BI, and LLM-backed analytics, say hello.