@sky_vst

Cloud, data geek and a runner. Cloud Advocate, Databases @ #GoogleCloud running in any weather. All opinions are my own. @sky_vst@hachyderm.io

 Ottawa,ON
blog.gleb.ca
Joined June 2009
  • Tweets2.3K
  • Following194
  • Followers556
496Media
Helping as a table captain on the #AgentsForImpact hackathon in Toronto. Great energy here. And it is really interesting how the mindset has to change to fully embrace AI coding tools. @googledevs #TechForGood
1
1
40
Lisbon marathon- done. 3:35:49 - not my best time but I take it. Great scenery course - highly recommended.
2
80
On the start of the Lisbon marathon - so many expensive shoes around! @LisbonMarathon
1
7
301
That's a great news! Going to adapt my codelab for the new embedding model. I will be presenting the lab on the Code Vipassana #16 on Oct 24 - register here codevipassana.dev @googledevs #EmbeddingGemma #PostgreSQL
Sundar Pichai
@sundarpichaiOct 6
Introducing EmbeddingGemma 2, a new open multimodal model that sets the standard for on-device efficiency. - our first open, natively multimodal embedding model - handles text, code, image, video, and audio tasks within a lightweight, modular 740M parameter form factor - ideal for offline, privacy-first RAG when paired with Gemma 4 - outperforms some specialist models more than twice its size Weights available now on Hugging Face.
54
Have you tried already TypeSafe AI's Jev model in AlloyDB? AlloyDB AI.* functions flexible enough to work with the model almost out of box and support batching approach which makes it really fast. I've written a short blog post on how to start using it. bit.ly/4iV3DGL

Sub-second semantic filtering in PostgreSQL: Integrating the ultra-fast Jev model with AlloyDB

Introduction

medium.com
1
81
Following the progress of the @SpaceX starship 14th launch. It is really incredible how much progress have been made. And videos from the ship looks like a video game.
1
71
How much overhead and added response time does it cost to store your vector embedding in a dedicated table in PostgreSQL? Can your application accept an additional 0.1 or 0.2 ms? Wrote a blog post checking the overhead. @googledevs #PostgreSQL #Vectors bit.ly/4ji31uQ

Inline vs. Separate Tables for Vectors in Postgres: Measuring the Join Overhead

Testing semantic search, filtered queries, and multi-table joins in AlloyDB to measure the true cost of decoupling your vectors.

medium.com
1
97
That's really cool! Test TPUs in space and see how it survives the radiation and temperature.
Sundar Pichai
@sundarpichaiSep 24
Can our TPUs survive and operate in space? Well, we're going to find out. Project Suncatcher is hitching a ride aboard @SpaceX's Transporter-18 mission, testing a prototype satellite built in partnership with @planet One small step for TPUs....
2
108
Great news - Bring Your Kid (Model) to Work in Google Antigravity. Now use your local Gemma for reasoning without going to the service endpoint.
N. Taylor Mullen
@ntaylormullenSep 23
Antigravity SDK September 21st 2026 weekly update for v0.1.18: • Introducing Support for Local AI Models in the Antigravity SDK: Run private, completely offline agentic workflows and hybrid cloud-local orchestrations powered by Gemma 4 26B A4B on Google AI Edge's or local OpenAI-compatible servers (for Ollama, LM Studio, and vLLM): - developers.googleblog.com/in… • Default Task Scheduling: Agents can now schedule deferred workflows across conversation turns using BuiltinTools.SCHEDULE in default toolsets. • Custom Subagent Models: Route specialized subagent tasks to dedicated model tiers by specifying SubagentConfig.model. github.com/google-antigravit… 🧵
1
1
323
Gleb Otochkin retweeted
N. Taylor Mullen
@ntaylormullenSep 23
Antigravity SDK September 21st 2026 weekly update for v0.1.18: • Introducing Support for Local AI Models in the Antigravity SDK: Run private, completely offline agentic workflows and hybrid cloud-local orchestrations powered by Gemma 4 26B A4B on Google AI Edge's or local OpenAI-compatible servers (for Ollama, LM Studio, and vLLM): - developers.googleblog.com/in… • Default Task Scheduling: Agents can now schedule deferred workflows across conversation turns using BuiltinTools.SCHEDULE in default toolsets. • Custom Subagent Models: Route specialized subagent tasks to dedicated model tiers by specifying SubagentConfig.model. github.com/google-antigravit… 🧵
12
4
3
58
4,336
Have you thought about how to manage your embeddings when you upgrade the model? The tables layout in your PostgreSQL database in some cases may lead to tables and index bloating. medium.com/google-cloud/ba6e… @googledevs #PostgreSQL #AlloyDB #CloudSQL

Embedding versions management, TOAST and bloating in PostgreSQL

If you are working with vector embeddings you probably already know the AI models including embedding models have their own lifecycle. When…

medium.com
85
Join us tomorrow on Database Maven event to discuss Agents, MCP and Databases. We are going to touch all three of them - Agents, how they work with MCP and how they use databases - and of course NL2SQL - we also should have time for Q&A at the end. luma.com/adurpjg0

Agents and MCP for Postgres · Zoom · Luma

AI agents need more than just prompts - they work with tools and data. This presentation explores the evolution of agentic workflows and the specific…

luma.com
1
81
I think Gemini 3.8 Flash is really good. It feels first that it is very close to 3.7 but then you start to notice some differences and quality of output on iterative tasks. I am switching all my agents to it. Now playing with different NL2SQL stuff. @googleaidevs
2
99
What do you think about running AlloyDB Omni on the same Kubernetes cluster with two open Gemma models? GemmaEmbedding to build your vectors and Gemma 4 to run all AI functions. Full control and you don't count tokens anymore. Try it out - goo.gle/4qTb9Cb @googledevs

AlloyDB Omni and EmbeddingGemma with Gemma 4 on Kubernetes  |  Google Codelabs

Learn how to deploy AlloyDB Omni and open embedding model and open LLM on GKE, register the models in AlloyDB Omni, and run vector search and LLM based queries directly from SQL.

codelabs.developers.google.com
5
323
I know somebody would say - "For a man with a hammer everything looks like a nail". But you can do a lot with Cloud SQL and AlloyDB AI integration. You can evaluate new models using your own data and criteria for example. goo.gle/4i3usrK @googledevs #AlloyDB #Gemini

Evaluating AI models with SQL in Google Cloud databases

Introduction

medium.com
94
You don’t need a title change to build apps. If you have an idea and curiosity, you can turn prompts into live applications with AI Studio + Google Cloud. Explore our new collection of guides covering web apps, games, spatial AI tools, and mobile apps → goo.gle/builders
9
55
13
236
70,791
That's really cool. I was always in opinion that the Developer Knowledge MCP was not appreciated enough. Now is even easier than ever - just run a gcloud - of course the biggest AlloyDB instance is 128 CPU.
Richard Seroter
@rseroterAug 24
You shouldn't have to hunt for which docs have the answer you want. Our Developer Knowledge API/MCP pulls latest from 19 different @Google doc sets. And we added it to the @googlecloud CLI. Get answers on Gemini, Android, Cloud, Chrome, Flutter, etc. docs.cloud.google.com/sdk/gc…
1
4
1,285
Spent some time this weekend working on a small project using VSCode with Google Antigravity extension. It was not too bad after all. The Gemini 3.7 Flash is quite good - especially when you know what to ask. @antigravity @googledevs
3
135
Join me at the Postgres Summit in NYC Sept 30 - Oct 2. For all who love and work with Postgres. I will have a talk "Securing the Data Layer for AI Agents and MCP Servers" on Friday Sept 2. Here is discount code -OTOCHKIN50 #PostgreSQL #AIAgents #MCP 2026.postgressummit.us/

Postgres Summit US 2026

A 3-day PostgreSQL community conference in New York City, USA.

2026.postgressummit.us
1
67
I used to know a guy who was wearing glasses just to look more geeky - maybe I should do that too.
78
Load more