Work
Analytics platform, built from zero
Sole data hire at a B2B proptech SaaS: ingestion, a dbt warehouse, a metric layer, four dashboards, alerting and predictive models. One person, end to end.
- 150+
- metrics, each defined once in git
- 100+
- dbt models, tests gate every build
- 3
- ML models: risk, churn, payments
Skriptly, AI transcription
Transcription and dictation I built and run solo. I rebuilt the engine twice: a hosted API, then open-source models on GPUs, then Soniox. Better accuracy, 4× cheaper.
- 60+
- languages, auto-detected
- 2.4%
- word error rate, down from 3.7%
- 4×
- cheaper per hour than self-hosted GPUs
About
I'm a Data & AI engineer in Prague. I joined a proptech platform as a product analytics intern in 2025, and I'm now the only person there who owns data and AI: the warehouse, dashboards, ML prototypes for building risk and churn, an MCP server over the metric layer, and analyst agents that answer questions and watch the metrics around the clock.
Next to that I build my own product. Skriptly is an AI transcription tool I started in May 2026 to turn my own work calls into clean notes. I run it solo: speech-to-text in 60+ languages, speaker recognition, summaries and action items, and a Windows app for dictation and call recording.
I work AI-native. I decide what to build and how it should behave, then use AI tooling to ship it fast, which is how one person can run a product and a data platform at the same time. What I care about most is the part that makes AI actually useful: clean data, clear metric definitions and a way to check the answers.
- 2025 — now Proptech SaaS · Data & AI Engineer, sole data hire
- 2026 — now Skriptly · Founder, AI transcription
Projects
-
lease-validator — evidence or nothing
2026An agentic RAG engine that checks lease agreements against YAML rules and refuses to say anything it cannot quote the document for.
0.976 violation F1 (baseline 0.692) · 0.960 F1 on adversarial docs (baseline 0.061) · 100% violation recall
-
metrics-mcp — a semantic layer an agent has to go through
2026Metrics defined once in YAML, SQL built from dbt, and every number an agent states checked against the result it came from before the sentence is allowed out.
11 metrics, 43 column refs checked · 100% invented numbers caught · 90% wrong-month claims caught
Writing
-
A check that always passes scores 100%
Nov 2026I built a guard that verifies every number an LLM states against the query it came from. Its score was perfect, and it was broken. How a deliberately lying stub found the bug.
-
I audited a week of our Slack alerts and most of them were noise
May 2026An alert bot everyone had learned to scroll past. What a week of its own messages showed, and the four changes that cut the volume by about two thirds.
Contact
I like talking with people building applied AI, with investors who follow the space, and with engineers who want to compare notes on LLM apps and data stacks. Open to data and AI engineering roles, and to contract work for early-stage teams.