My work, my passions, my interests.

Welcome! Here I want to share what I actually work on — my projects, my writing, the things I'm curious about. I've been in analytics for 15+ years, since back when data science was still called statistics. These days I'm mostly researching and experimenting with AI — at work, and in my own projects.

15+Years in analytics & AI
4Agents built
63National parks — the quest
14Countries traveled
Connect on LinkedIn Read on Medium
Anastasiia at DeepLearning.AI's AI Dev 26 conference
AI DEV 26STAYING CURRENT
A fjord in Norway, one stop on the travels
10 COUNTRIESAND COUNTING
Agent handoff design for the closet & wardrobe recommender
WARDROBE AGENTHANDOFF DESIGN
01
AI experiments

Three things you'll find here about my AI experiments.

Build
Prove
Translate
01

Multi-agent systems, from scratch

Specialist agents over a unified data layer, with persistent memory so they don't rebuild context every run. I design the architecture, then stay to make it reliable.

4 agents built
02

Evaluation before deployment

An LLM output nobody validated is a hypothesis, not an answer. I build the eval harnesses — accuracy, tone, completeness, hallucination rate — that make shipping defensible.

4 eval dimensions
03

Modelling into decisions

Segmentation, forecasting, and risk models are only worth what someone does with them. I work backwards from the decision a business actually needs to make.

15+ years analytics
All projects →
02
Beyond work

Not quite hobbies. Or maybe they are.

63 parks
one quest
The quest

All 63 US national parks

A finite list, a long timeline, and a spreadsheet color-coded more than it should be.

— of 63 visited
Everywhere else

Travel outside the US

Trips across Europe, Scandinavia, Asia, Africa, and Latin America — same instinct as the parks list, just without a finish line.

— countries visited
See beyond work →
03
From the writing

Ideas I'm learning, then writing about.

2 pieces
more on
Medium
Prompt engineering · System design

Prompt Engineering Is Not Engineering. It's System Design.

A better prompt can fix an ambiguous instruction. It can't fix missing context, an unvalidated tool interface, or a role carrying too much responsibility.

Read → Multi-agent · Failure modes

Where does your multi-agent system actually break?

The failures that matter in multi-agent systems aren’t crashes — they’re the handoffs, the shared state, and the moments one agent quietly contradicts another.

Read →
Read the writing → Follow on Medium
04
AI Dev 26

How I keep up.

Eight themes, dozens of sessions, and a few things worth remembering — my field report from DeepLearning.AI’s AI Dev 26 conference.

Read the field report →