Pineapple on pizza

No data available

Feature ownership
  • Data models
    Views, materialized views, orchestration, and lineage.
  • Data quality checks
  • Endpoints
  • Semantic layer and data catalog
  • Warehouse properties
    Syncs tables and materialized views to person or group properties.
Members

Roadmap

Recently shipped

Clearer view failure digests

Materialized-view failure emails now separate views that need action from newly failing and ongoing failures. Each row includes the failure streak and, when relevant, what to do next.

Goals

Q4 2026 objectives

Objective 1: Semantic layer (owner: Sinan Kucukkoseler)

Motivation: The biggest bottleneck for the semantic layer is getting metrics and definitions into it. Users also can't see how much of their analysis it covers, so they don't know what to add or how far to trust it.

What we'll ship:

  • A coverage mechanism that shows how often metrics are used, what they cover, and what's missing
  • An Ossie-like spec for defining the semantic layer
  • A clear split between the semantic layer and the context layer, and what specialized agents on top of them look like
  • Exploration: a standalone semantic layer, depending on demand
  • Answer the question: how do we increase trust in data across all users?

Objective 2: Self-driving warehouse (owner: Thiago Rocha Salvatore)

Motivation: Most people don't know which views, materialized views, Endpoints, or metrics they need. We can see what they query, so we should act as their data engineer and propose those changes to them.

What we'll ship:

  • Recommendations on what to materialize, certify, and deprecate, and how to configure materializations
  • Metric proposals based on what users actually query
  • Query optimization and fixes for slow views and Endpoints
  • Integration with the Inbox
  • Recommendations delivered through the version control workflow for data modeling

Objective 3: Modeling on the Trino warehouse (owner: Jovan Sakovic)

Motivation: We need to move data modeling workloads off ClickHouse and onto the Trino warehouse, so we can charge for modeling and give customers compute that scales with them.

What we'll ship:

  • PostHog's own revenue dbt project running on data modeling
  • All data models running on Trino
  • Faster modeling jobs on Trino
  • A defined transition from ClickHouse to Trino, including limits on ClickHouse and benchmarks against Snowflake and BigQuery
  • Revenue tracking for modeling downstream of the Trino warehouse

Objective 4: Version control (owner: Jovan Sakovic)

Motivation: dbt is ahead of us on trust. To compete, data teams need versioned models and tests they can review and roll back, the same way they work today.

What we'll ship:

  • A git-backed, dbt-style data modeling workflow for agents and humans, including data quality checks
  • A CLI to manage the project
  • A way to take over your PostHog data modeling project and host it yourself
  • Integration with Trino
  • PostHog's own revenue dbt project managed this way

Objective 5: Product metrics for the data stack are properly modeled (owner: Anna Szell)

Motivation: Currently, product metrics are all over the place, and there's no structure around how metrics are defined. Also, this is a good way to dogfood all the improvements we are making to the modeling product.

What we'll ship:

  • All key metrics are modeled, have their own DAG, version-controlled, and tested
  • Write skills specifically for PMs to easily reproduce these for their own teams/products
  • Write blog post / handbook article to share more widely and document how we do data modeling for product metrics

Stretch: Python in the modeling lineage (owner: Sinan Kucukkoseler)

What we'll ship:

  • Run a Python script as a step in a larger modeling lineage

Handbook

Features

Slack channel

#team-data-modeling