Algirdas Rumšas · Lithuania · remote-first
Analytics & AI engineer for governed data platforms—and the AI layer on top.
10+ years across the full data stack. I build medallion lakehouses on Microsoft Fabric, analytics engineering with dbt and BigQuery, and LLM workflows grounded in semantic models and RAG. Open to full-time or substantive part-time (from ~2 weeks/month) engagements.
Past and current collaborators include OrderYOYO, Whatagraph, Cobiro, Microsoft Fabric, and BigQuery · dbt.
Selected work
A few high-signal pieces—products and one representative engagement—not an exhaustive list.
- Product
Forbi
- Problem
- Teams live in Power BI but still burn time hunting metrics and arguing about definitions.
- Approach
- An AI agent on the semantic model: metadata is vectorised, STAR-shaped breakdowns are mapped, and answers are forced through governed metrics—not guessed DAX.
- Result
- Faster self-serve analytics with hallucination risk bounded to metric choice, not calculation.
- Product
ntxt
- Problem
- AI tools forget context between sessions; black-box memory is hard to trust or steer.
- Approach
- A persistent, queryable context graph exposed to assistants via MCP—visible, portable memory you control.
- Result
- Shared memory across tools without losing auditability of what the model can see.
- Past role · 2020–2024
Whatagraph
- Problem
- The business needed reliable core metrics (MRR, retention, activation) and models sales and CS could act on.
- Approach
- dbt + BigQuery analytics, propensity and upsell models, instrumentation standards with engineering—and an outlier detector that shipped as a product feature.
- Result
- Monitoring leadership could trust, ML that moved from internal insight to customer-facing value.
- Consulting · Microsoft Fabric
OrderYOYO (return engagement)
- Problem
- Fraudulent chargebacks spiked; the org needed end-to-end delivery inside a large corporate structure.
- Approach
- Full pipeline on Fabric and Python notebooks: ingestion, transformation, Power BI, stakeholder-ready narrative—and detection plus dispute response for Braintree/PayPal abuse patterns.
- Result
- Chargebacks down from roughly £30k/month to about £5k/month.
Get in touch for a fuller CV or a focused conversation on Fabric, BI, or AI-on-analytics.
About
I operate as Fulljoin and gravitate toward 0-to-1 data situations: the first warehouse, the first trustworthy metric layer, the first time AI is allowed near production numbers.
I care about explainability and pragmatism over black-box models—treating data code like production software (SQL, Python, CI/CD, clear ownership).
Contact
If you need a senior IC who can span lakehouse, BI, and LLM-backed analytics, say hello.