2026-09-10 –, Unconference
Dashboards tell you what happened. They rarely tell you when it matters.
In this talk, we present a Python-first approach to building active KPI systems on top of the modern data stack. Using a semantic model to define metrics, dimensions, and business logic, we move beyond static dashboards toward event-driven data workflows.
We’ll show how to define KPIs as code using typed schemas, compute them using DuckDB, and attach intelligent alerting that triggers when meaningful changes occur. Not just threshold breaches. Instead of brittle “if > X then alert” logic, we introduce context-aware rules, anomaly detection, and dependency aware KPI evaluation.
We’ll also cover a lightweight Python-based alerting framework that integrates with common tools (Slack, email, APIs), enabling self-service data monitoring.
Attendees will learn how to turn passive dashboards into operational data system. And yes, while it could be enhanced with, in its core this works entirely without LLMs.
Example code and a minimal alerting framework will be shared.
Topic & Relevance
Most organizations rely on dashboards for monitoring KPIs, resulting in delayed reactions and missed signals. Meanwhile, data teams struggle with alert fatigue and brittle threshold systems.
This talk introduces a Python-based, event-driven KPI framework that transforms metrics into actionable signals.
Technical Depth
KPI definitions using Pydantic / typed schemas
Execution using DuckDB
Rule-based and statistical alerting
Dependency graphs between KPIs
Alert delivery via Slack/webhooks
Anomaly detection (lightweight stats)
Target Audience
Data engineers
Analytics engineers
Data scientists working with metrics and monitoring
Level:
Intermediate
Audience Takeaways
How to define KPIs as code (not dashboards)
How to build alerting systems that reduce noise
How to structure dependency-aware KPI evaluation
How to enable self-service monitoring for teams
Required Background
Python
Basic data analysis (Pandas/SQL)
Familiarity with KPIs or business metrics
Talk Type & Approach
Practical, system design oriented
Code examples + architecture
Real-world patterns (not theoretical)
Outline (30 min)
Introduction why dashboards fail (5 min): Passive vs active data systems
KPI Modelling (8 min): Metrics as code, semantic definitions
Execution Layer (5 min): Computing KPIs with DuckDB (or Pandas/Polars)
Alerting System (8 min): Rules, anomaly detection, dependencies
System Demo / Example Flow (3 min): From KPI change → alert → action
Takeaways (1 min): Principles for operational data systems
Building Python-first data systems at the intersection of modern data stacks and corporate intelligence, with a focus on integration, modelling, and reliability.