by Jacek Migdal
How often do you attend a tech conference in Freemasonry temple that features staged protest and piano performance? Data Council 2025 delivered exactly that, along with the hidden gem talks that later translated into nearly $1B in combined investment and acquisitions.
Specifically, ClickHouse ($350M funding), Harvey ($300M funding), and CrunchyData (acquired for $250M) made headlines soon after the conference. Previous editions showcased Iceberg before it was acquired for $2B and WarpStream before its acquisition by Confluent.
Let’s recap our favorite 2025 talks, share insights, and highlight the aesthetic aspects of the event.
Venue: the Masonic Temple of Data
Conference memories often fade quickly, blending into repetitive patterns of booths, badges, and generic venues – typical “fast fashion” experiences. At Data Council 2025, the Oakland Scottish Rite Center stood out distinctly from typical convention centers. It's a historical Freemasonry temple, rich with symbolic details.
Rather than plunge into conspiracy theories, the conference offered deep insights into the data industry. Attendees were greeted by a staged "Free the Analyst" protest, breakfast burritos, and an opening talk by Pete Soderling, the conference founder. Pete even performed an impromptu piano piece, "Hallucinate-ya", a playful LLM-themed twist on "Hallelujah."
Atmosphere aside, the ample space for hallway conversations and speaker office hours enhanced the experience, making these informal interactions as valuable as the formal sessions – especially among founders (including Y Combinator alumni) and data infrastructure builders.
Favorite talks
Here are my top picks, emphasizing deep technical content. You can also view the full schedule and access all recordings. I renamed the talk titles to be more descriptive than catchy.
The history of DuckDB
Ducks weren't only hatching at nearby Meridian Lake – they were also featured at the temple of data. DuckDB co-creator Hannes Mühleisen explained its emergence as a reaction to overly complex and expensive analytical tools circa 2015. Inspired by SQLite's simplicity, DuckDB adopted a single-node, vectorized design without GPU dependencies, excelling from browsers to cloud servers. Their innovation continues with DuckLake, challenging Apache Iceberg.
Hidden gem: Harvey’s legal agency
This talk was criminally under-viewed, with only 90 views by June 30. Harvey’s launch underscores the need for specialized AI infrastructure and effective human-computer interaction in legal tech. Niko Grupen's talk covered managing complex data scenarios, rigorous evaluation methods, and hierarchical agent workflows tailored to legal use cases.
Architecture of Turbobuffer (used by Notion, Cursor and Linear)
This joint talk from Notion and Turbopuffer showcased vector search on top of object storage. Notion demonstrated the practical benefits of Turbopuffer’s indexing innovations, emphasizing cost-efficiency and speed. Turbopuffer detailed its technical optimizations, including SSD caching, minimal roundtrips to object storage, and clustered indexing.
Applied LLMs in data
Hadley Wickham, the creator of the stunning charting library ggplot2, addressed the paradox of LLMs: they are great at creativity, yet weak in computational precision. He demonstrated valuable use cases like idea generation, code automation, and data extraction from unstructured sources, advocating for human-AI collaboration, or "data science centaurs."
SQL compile-time innovation
Elias DeFaria, co-founder of SDF (acquired by dbt), presented "Fusion," a fast, local SQL engine written in Rust. It offers interactive feedback, supports multi-dialect SQL, and provides detailed lineage metadata.
Instant SQL preview
MotherDuck launched "Instant SQL," introduced by Hamilton Ulmer. Inspired by Digital Audio Workstations, it provides real-time previews of query results after each keystroke, significantly improving SQL debugging and observability on DuckDB.
AI Launchpad
Beyond the deep dives, the conference also provided a platform for the six startups that gave pitches in the AI Launchpad session:
Quesma Charts: AI-powered charting (admittedly biased).
Nao: Cursor for Data Science.
Mooncake: Real-time analytics on Postgres tables.
TopK: Document database with native multi-vector, keyword, and faceted search (raised $5.5 million since the conference)
Guilde: AI assistant for data engineers.
Tower: Serverless Python platform.
Other highlights
The best swag award goes to DltHub for their data tarot cards. Data analytics is prone to legacy and overall complex setups, so let's have some laughs instead of cursing.
The second place goes to the Firebolt's red dragon, your new desk mascot. Not only is the team on fire with a new FireScale benchmark and the free embedded DB Firebolt Core, but they also have a performant data warehouse and cool swag.
Final thoughts
So, what's the final verdict? Data Council remains an excellent conference for builders. It delivers deep content on relevant trends while steering clear of typical industry hype. The event deserves broader attention, especially among engineering leads and principal engineers. However, it's currently less effective for direct user acquisition – a common trend in the post-ZIRP era, where conferences face challenges as a go-to-market channel.
