Explore vector-powered Postgres for AI with Gleb Otochkin at PG DATA 2026 on June 4!

In โ€œVector data in Postgres: Size, TOAST, Filters and Performance,โ€ Gleb dives into how PostgreSQL handles vector data for AI-driven applications ๐Ÿ˜

Join us: https://2026.pg-data.org/

#PGData #PGData2026 #PostgreSQL #Postgres #AI #VectorSearch #Database #OpenSource #PerformanceTuning #DataEngineering

What Matters in Production RAG

์ด ๊ธ€์€ ํ”„๋กœ๋•์…˜ ํ™˜๊ฒฝ์—์„œ RAG(Retrieval-Augmented Generation) ์‹œ์Šคํ…œ ๊ตฌ์ถ• ์‹œ ํ”ํžˆ ๊ฐ„๊ณผ๋˜๋Š” ํ•ต์‹ฌ ๊ธฐ์ˆ ์  ๋„์ „๊ณผ ํ•ด๊ฒฐ์ฑ…์„ ๋‹ค๋ฃฌ๋‹ค. ๋ฌธ์„œ ์ฒญํ‚น, ์ž„๋ฒ ๋”ฉ ๋ชจ๋ธ ๊ณ ์ • ๋ฌธ์ œ, ์ธ๋ฑ์Šค ๊ฐฑ์‹ ๊ณผ ๋ฒ„์ „ ๊ด€๋ฆฌ, ๋ถˆํ•„์š”ํ•œ ์žฌ์ž„๋ฒ ๋”ฉ ๋ฐฉ์ง€, ๋ฌด์ค‘๋‹จ ์ธ๋ฑ์Šค ์—…๋ฐ์ดํŠธ ๋“ฑ ์‹ค๋ฌด์—์„œ ๋ฐ˜๋“œ์‹œ ๊ณ ๋ คํ•ด์•ผ ํ•  ์‚ฌํ•ญ๋“ค์„ ์ƒ์„ธํžˆ ์„ค๋ช…ํ•œ๋‹ค. ํŠนํžˆ ๋Œ€๊ทœ๋ชจ ๋ฌธ์„œ ์ง‘ํ•ฉ์„ ๋‹ค๋ฃจ๋Š” AI ์„œ๋น„์Šค ๊ฐœ๋ฐœ์ž์—๊ฒŒ ์œ ์šฉํ•œ ์ธ์‚ฌ์ดํŠธ๋ฅผ ์ œ๊ณตํ•œ๋‹ค.

https://arpitbhayani.me/blogs/rag-production/

#rag #retrievalaugmentedgeneration #vectorsearch #embedding #aiinfrastructure

What Matters in Production RAG

Most of us build RAG the same way: follow a tutorial that embeds a handful of PDFs, stores the vectors in a local Chroma instance, and chains everything together with LangChain (if that's still a thing). The demo works. The answer looks reasonable. Then you take it to production and it falls apart in quiet, hard-to-diagnose ways.

Arpit Bhayani

StyloBot free day as I ran myself ragged trying to get it going in my free time (very little of which I HAD finishing up 2x contracts!).

Biggest win is dropping the ONNX dependency.

Earlier versions used ONNX embeddings as a shortcut: turn a client signature into a vector and compare it.

It worked, but it was never quite the right abstraction. Embeddings are built for language. StyloBotโ€™s inputs are behavioural structures.

The new version defines that behavioural vector space directly. Requests, sessions, browsers, bots, scrapers, and odd clients are placed into a real StyloBot-native space. The system ships with archetype centroids, then adapts those centroids to the actual traffic it sees.

So instead of asking a model what a client 'means', StyloBot learns what your traffic looks like.

StyloBot is REALLY a conceptually unfolded ML model so it sort of trains itself on real traffic around centroids and updates as it goes. It's ODD.

Now out in Release Candidate https://github.com/scottgal/stylobot/releases

Plan is still for full release June 1st but the FOSS client MAY reach RTM quality before that (lots of manual testing!)

#BotDetection #CyberSecurity #DotNet #SQLiteVec #VectorSearch #BehaviouralInference #AIInfrastructure #OpenSource

AionDB: PostgreSQL-compatible SQL, graph, and vector database in Rust

AionDB๋Š” Rust๋กœ ๊ฐœ๋ฐœ๋œ PostgreSQL ํ˜ธํ™˜ SQL, ๊ทธ๋ž˜ํ”„, ๋ฒกํ„ฐ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋กœ, ๊ด€๊ณ„ํ˜• ๋ฐ์ดํ„ฐ, ๊ทธ๋ž˜ํ”„ ๊ด€๊ณ„, ๋ฒกํ„ฐ ๊ฒ€์ƒ‰์„ ํ•˜๋‚˜์˜ ์—”์ง„์—์„œ ์ง€์›ํ•ฉ๋‹ˆ๋‹ค. pgwire ํ”„๋กœํ† ์ฝœ๊ณผ ์—ฌ๋Ÿฌ ๊ฒ€์ฆ๋œ ORM์„ ํ†ตํ•ด ๊ธฐ์กด PostgreSQL ์ƒํƒœ๊ณ„ ๋„๊ตฌ์™€ ํ˜ธํ™˜๋˜๋ฉฐ, SQL๊ณผ Cypher ์Šคํƒ€์ผ ์ฟผ๋ฆฌ๋ฅผ ๋™์‹œ์— ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์„ฑ๋Šฅ ๋ฉด์—์„œ SurrealDB ๋Œ€๋น„ ๊ฒฝ์Ÿ๋ ฅ ์žˆ๋Š” ๊ฒฐ๊ณผ๋ฅผ ๋ณด์ด๋‚˜, PostgreSQL ๋Œ€์ฒด๋‚˜ ์™„์ „ํ•œ ๋ถ„์‚ฐ ํด๋Ÿฌ์Šคํ„ฐ ๊ธฐ๋Šฅ์€ ์•„์ง ์ง€์›ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ํ˜„์žฌ ์•ŒํŒŒ ๋ฒ„์ „์œผ๋กœ, ์‹ค์ œ ์ ์šฉ ์‹œ ๊ธฐ๋Šฅ๋ณ„ ๊ฒ€์ฆ์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.

https://aiondb.xyz/

#rust #database #postgresql #vectorsearch #graphdb

Home - aiondb

How to Build Vector Search from Scratch in Python

์ด ๊ธ€์€ Python๊ณผ NumPy๋งŒ ์‚ฌ์šฉํ•ด ๋ฒกํ„ฐ ๊ฒ€์ƒ‰ ์—”์ง„์„ ์ฒ˜์Œ๋ถ€ํ„ฐ ๊ตฌํ˜„ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์ƒ์„ธํžˆ ์„ค๋ช…ํ•œ๋‹ค. ํ…์ŠคํŠธ๋ฅผ ๊ณ ์ฐจ์› ์ž„๋ฒ ๋”ฉ ๋ฒกํ„ฐ๋กœ ๋ณ€ํ™˜ํ•ด ์ฝ”์‚ฌ์ธ ์œ ์‚ฌ๋„๋กœ ์˜๋ฏธ์  ๊ทผ์ ‘์„ฑ์„ ์ธก์ •ํ•˜๋Š” ๋ฒกํ„ฐ ๊ฒ€์ƒ‰์˜ ๊ธฐ๋ณธ ์›๋ฆฌ๋ฅผ ๋‹ค๋ฃจ๋ฉฐ, ๊ฐ„๋‹จํ•œ ์ƒํ’ˆ ์„ค๋ช… ๋ฐ์ดํ„ฐ์…‹์„ ํ™œ์šฉํ•ด ์ž„๋ฒ ๋”ฉ ์ƒ์„ฑ, ์ •๊ทœํ™”, ์ธ๋ฑ์‹ฑ, ๊ฒ€์ƒ‰ ์ฟผ๋ฆฌ ์ฒ˜๋ฆฌ ๊ณผ์ •์„ ๋‹จ๊ณ„๋ณ„๋กœ ๋ณด์—ฌ์ค€๋‹ค. ๋˜ํ•œ PCA๋ฅผ ์ด์šฉํ•ด ์ž„๋ฒ ๋”ฉ ๊ณต๊ฐ„์„ 2์ฐจ์›์œผ๋กœ ์‹œ๊ฐํ™”ํ•ด ํด๋Ÿฌ์Šคํ„ฐ ๊ตฌ์กฐ์™€ ์ฟผ๋ฆฌ ๋ฒกํ„ฐ์˜ ์œ„์น˜๋ฅผ ์ง๊ด€์ ์œผ๋กœ ์ดํ•ดํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•œ๋‹ค. ๋ฒกํ„ฐ ๊ฒ€์ƒ‰์˜ ํ•ต์‹ฌ ๊ฐœ๋…๊ณผ ๊ตฌํ˜„ ์›๋ฆฌ๋ฅผ ์ดํ•ดํ•˜๊ณ ์ž ํ•˜๋Š” AI ๊ฐœ๋ฐœ์ž์—๊ฒŒ ์‹ค์šฉ์ ์ธ ์ž…๋ฌธ ์ž๋ฃŒ๋‹ค.

https://www.kdnuggets.com/how-to-build-vector-search-from-scratch-in-python

#vectorsearch #python #embedding #cosinesimilarity #pca

How to Build Vector Search from Scratch in Python

Learn how to build a vector search engine from scratch in Python with embeddings, similarity scoring, and basic retrieval logic.

KDnuggets

Deep Dive: Vector Search in Hermes Memory

SQLite vector search implementation details.

#memory #vectorsearch #sqlite

Building an AI Agent with Persistent Memory: A Technical Deep Dive

A technical look at how Hermes Agent implements cross-session persistent memory using SQLite vector search and knowledge graphs.

#ai #agents #memory #vectorsearch #opensource

Use Redis with SQL

Redis์—์„œ SQL๊ณผ ์œ ์‚ฌํ•œ ์ฟผ๋ฆฌ๋ฅผ ์‹คํ–‰ํ•  ์ˆ˜ ์žˆ๋Š” sql-redis ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ๊ฐ€ PyPi์— ๊ณต๊ฐœ๋˜์—ˆ๋‹ค. ์ด ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ๋Š” SQLQuery ํด๋ž˜์Šค๋ฅผ ํ†ตํ•ด SQL ๋ฌธ์„ Redis์˜ FT.SEARCH, FT.AGGREGATE ๋ช…๋ น์–ด๋กœ ๋ณ€ํ™˜ํ•˜์—ฌ Redis์˜ ๋น ๋ฅธ ์†๋„๋กœ ์ฟผ๋ฆฌ๋ฅผ ์ˆ˜ํ–‰ํ•œ๋‹ค. ๊ธฐ๋ณธ์ ์ธ SELECT ๋ฌธ๋ฟ ์•„๋‹ˆ๋ผ ์ง‘๊ณ„, ์ „์ฒด ํ…์ŠคํŠธ ๊ฒ€์ƒ‰, ๋ฒกํ„ฐ ๊ฒ€์ƒ‰, ์ง€๋ฆฌ ๊ณต๊ฐ„ ์ฟผ๋ฆฌ, ๋น„๋™๊ธฐ ์‹คํ–‰๋„ ์ง€์›ํ•˜๋ฉฐ, LLM ์—†์ด deterministicํ•˜๊ฒŒ ๋™์ž‘ํ•œ๋‹ค. RedisVL ํŒจํ‚ค์ง€์— ํฌํ•จ๋˜์–ด ์žˆ์–ด ๊ฐ„๋‹จํ•œ ์„ค์น˜์™€ ์‚ฌ์šฉ๋ฒ•์œผ๋กœ Redis ์ธ๋ฑ์Šค์— ์นœ์ˆ™ํ•œ SQL ๋ฌธ๋ฒ•์œผ๋กœ ์ ‘๊ทผ ๊ฐ€๋Šฅํ•˜๋‹ค. AI ๊ฐœ๋ฐœ์ž๋“ค์ด Redis๋ฅผ ๋ฐ์ดํ„ฐ ์ €์žฅ์†Œ๋กœ ํ™œ์šฉํ•˜๋ฉด์„œ SQL ์นœํ™”์ ์ธ ์ฟผ๋ฆฌ ๊ฒฝํ—˜์„ ์–ป๊ธฐ์— ์œ ์šฉํ•˜๋‹ค.

https://redis.io/blog/use-redis-with-sql/

#redis #sql #vectorsearch #fulltextsearch #python

Use Redis with SQL | Redis

Developers love Redis. Unlock the full potential of the Redis database with Redis Enterprise and start building blazing fast apps.

Redis

Google for Developers (@googledevs)

Gemini Embedding 2๊ฐ€ Matryoshka Representation Learning(MRL)์„ ํ™œ์šฉํ•ด ์ž„๋ฒ ๋”ฉ ํšจ์œจ์„ ๋†’์ธ๋‹ค๋Š” ๋‚ด์šฉ์ž…๋‹ˆ๋‹ค. ๋ฒกํ„ฐ๋ฅผ ๋™์ ์œผ๋กœ ์ž˜๋ผ ๊ณ ์† ํ›„๋ณด ๋งค์นญ์„ ํ•˜๋ฉด์„œ๋„ ์ •๋ฐ€๋„๋ฅผ ์œ ์ง€ํ•˜๊ณ , ๋” ์ž‘์€ ์ €์žฅ์†Œ๋กœ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค ๋น„์šฉ๋„ ์ค„์ผ ์ˆ˜ ์žˆ๋‹ค๊ณ  ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค.

https://x.com/googledevs/status/2051773542513947092

#gemini #embeddings #mrl #vectorsearch #ai

Google for Developers (@googledevs) on X

Matryoshka dolls ๐Ÿช† = the key to AI efficiency. Gemini Embedding 2 leverages Matryoshka Representation Learning (MRL) so you can: ๐Ÿ”น Dynamically truncate vectors for high-speed candidate matching without losing precision ๐Ÿ”นSlash database costs by choosing a smaller storage

X (formerly Twitter)

#VirtualThreads arenโ€™t just a #Java hype feature. This article shows them powering agent calls safely in production-style #Microservicesโ€”with fallback + observability.

Steal the blueprint by @sibaspadhi: https://javapro.io/2026/01/22/java-25-genai-a-new-era-for-microservices-in-finance/

#SpringBoot #GenAI #Observability #VectorSearch