---
title: AI Night Tallinn by ClickHouse (October 30, 2025)
date: 2025-10-30
order: -20251030
tags: [ai, events, clickhouse, docker, агенты]
---

Visited AI meetup organized by Zoe Steinkamp from ClickHouse and Oleg Šelajev from Docker, Inc.

Zoe did a blazing fast introduction to ClickHouse. ClickHouse is an open-source, columnar OLAP database built for ultra-fast analytics on massive datasets. It powers real-time dashboards, AI data pipelines, and observability systems by efficiently storing and querying large-scale data. Recent developments include support for feature stores, vector search, and AI-assisted querying, enabling businesses to use curated internal data for machine learning and natural language SQL generation.

![](/ru/events/img/1761858534343.jpg)

My question was about scaling it. Apparently it can scale vertically and horizontally. We at Clarifai heavily rely on vector databases for AI solutions, so it is interesting for sure to benchmark - see some of my old benchmarks at https://lnkd.in/dQr4edD5

I also tried InfluxDB in my Gratheon project for storing IoT data, but it was a bit of an overkill and hard to maintain for such small scale for now. Maybe I should consider ClickHouse in the future..

<!--truncate-->

Next, Vitalii Lakusta did a pretty extensive overview of agentic AI landscape, bringing audience up to speed with how cool modern AI infrastructure is both for development and for other things, like debugging, documentation.
https://mastra.ai/book was mentioned as well as some tool by Carl Rannaberg that helps with Claude hooks (I lost the link though)

Here I was questioning context limits, agent orchestration and I wanted to also dive into humans becoming proxies when AI does development and reviews but we didn't have much time for Q&A.

Next Oleg Šelajev showed that Docker for Desktop actually has MCP configuration and you can declare and spin agents just like you do with docker-compose, using https://lnkd.in/dGPSFVkJ

![](/ru/events/img/1761858532831.jpg)

Here I was questioning agent-to-agent communication over network (A2A)

Lastly, Eero Koplimets from Katana Cloud Inventory presented a real-life data architecture issue and journey of migrating from Postgres with materialized views to ClickHouse, cutting down costs in 3x (or more?) and reducing data latency from 30 min to realtime (3s?) and how all of this was done using AI not just in analyzing SQL but also in writing all of the edge cases and tests, ensuring that customers do not get any inconsistencies.

Here I was questioning how results were compared if data was changing.

## Related

- [AWS Gen AI meetup (May 7, 2025)](/en/events/2025-05-07-aws-gen-ai-meetup/)
- [Decentralized AI event (April 28, 2025)](/en/events/2025-04-28-decentralized-ai-event/)
- [Tallinn AI meetup (December 13, 2024)](/en/events/2024-12-13-tallinn-ai-meetup/)