RESEARCH & FIELD NOTES
What we measure, published.
Benchmarks, failure taxonomies, and build notes from making AI answers trustworthy on open-source stacks. No jargon fog, receipts included.
Why Most AI Companies Don't Need Real-Time Data (And What They Actually Need)
The obsession with real-time data pipelines is costing companies millions. Here's how to figure out what latency you actually need.
Feb 10, 2026 · 6 min read
Building LLM Observability From Scratch: A Practical Guide
How to build a complete observability stack for your LLM applications without expensive third-party tools.
Feb 5, 2026 · 8 min read
5 Data Architecture Mistakes AI Startups Make (And How to Avoid Them)
Common pitfalls we see when AI companies try to build their data infrastructure, and the patterns that actually work.
Jan 28, 2026 · 7 min read
Airflow vs. Dagster vs. Mage: Choosing the Right Data Orchestration Tool
An honest comparison of popular data orchestration tools based on real project experience.
Jan 20, 2026 · 9 min read
Self-Hosted BI: Why We Recommend Superset and Metabase Over Paid Tools
Enterprise BI tools cost a fortune. Here's why self-hosted open-source alternatives are often the better choice.
Jan 12, 2026 · 5 min read
In progress
- We benchmarked AI's SQL accuracy on a deliberately messy database.
- A field guide to plausible wrong answers.
- What actually changed: raw score to fixed score.
- The open-source accuracy stack: dbt + Cube + governed access on Postgres and ClickHouse.
- The AI Data Accuracy Eval Kit is open source.
Find out what your AI actually scores.
One week. Thirty questions. A number instead of a hope.