cevian
@ceviantech.bsky.social
Tech lead for pgai, pgvectorscale, and all other AI things @ Timescale.
Postgres rocks!
Postgres rocks!
For the past few months, I've been working on teaching AI coding tools to write better Postgres code. It's a passion project born from frustration.
October 22, 2025 at 5:49 PM
For the past few months, I've been working on teaching AI coding tools to write better Postgres code. It's a passion project born from frustration.
I’ve been thinking about why LLMs are bad at text-to-sql.
Short answer: no context 🧵
Compared to developers, LLMs are more handicapped without context:
- They can’t read the application code.
- They can’t ask around.
- They don’t know where the docs are.
Short answer: no context 🧵
Compared to developers, LLMs are more handicapped without context:
- They can’t read the application code.
- They can’t ask around.
- They don’t know where the docs are.
August 21, 2025 at 8:36 PM
I’ve been thinking about why LLMs are bad at text-to-sql.
Short answer: no context 🧵
Compared to developers, LLMs are more handicapped without context:
- They can’t read the application code.
- They can’t ask around.
- They don’t know where the docs are.
Short answer: no context 🧵
Compared to developers, LLMs are more handicapped without context:
- They can’t read the application code.
- They can’t ask around.
- They don’t know where the docs are.
I wish people would analyze benchmarks with this kind of depth and insight more often.
An analysis of Search Benchmark, the Game
Adrien decided to run an own blog. This shares an overview over a well known search benchmark maintained by the fine folks at Tantivy and breaks down different benchmarks and their implementations & tradeoffs.
Adrien decided to run an own blog. This shares an overview over a well known search benchmark maintained by the fine folks at Tantivy and breaks down different benchmarks and their implementations & tradeoffs.
An analysis of Search Benchmark, the Game
“Search Benchmark, the Game” is maintained at https://github.com/quickwit-oss/search-benchmark-game by the Tantivy folks and published at https://tantivy-search.github.io/bench/. I don’t know the…
jpountz.github.io
May 19, 2025 at 1:57 PM
I wish people would analyze benchmarks with this kind of depth and insight more often.
🚫 Stop treating embedding generation like ETL. ✅ It’s indexing.
One of the most common misconceptions in the world of AI and LLMs: 👉 Vector embedding generation = ETL. Nope. It’s much more like building an index.
Conceptually, it's just like creating a full-text index on your data.
One of the most common misconceptions in the world of AI and LLMs: 👉 Vector embedding generation = ETL. Nope. It’s much more like building an index.
Conceptually, it's just like creating a full-text index on your data.
May 9, 2025 at 8:48 PM
🚫 Stop treating embedding generation like ETL. ✅ It’s indexing.
One of the most common misconceptions in the world of AI and LLMs: 👉 Vector embedding generation = ETL. Nope. It’s much more like building an index.
Conceptually, it's just like creating a full-text index on your data.
One of the most common misconceptions in the world of AI and LLMs: 👉 Vector embedding generation = ETL. Nope. It’s much more like building an index.
Conceptually, it's just like creating a full-text index on your data.
🐘PGAI VECTORIZER NOW WORKS WITH ANY POSTGRES DATABASE (incl Timescale Cloud, Amazon RDS, Supabase, Azure PostgreSQL, and more)
www.timescale.com/blog/pgai-ve...
www.timescale.com/blog/pgai-ve...
Pgai Vectorizer Now Works With Any Postgres Database
By popular demand, pgai Vectorizer is now a Python library and CLI that works with any self-hosted or managed Postgres database. See how we built it.
www.timescale.com
May 1, 2025 at 4:59 PM
🐘PGAI VECTORIZER NOW WORKS WITH ANY POSTGRES DATABASE (incl Timescale Cloud, Amazon RDS, Supabase, Azure PostgreSQL, and more)
www.timescale.com/blog/pgai-ve...
www.timescale.com/blog/pgai-ve...
Listening to our CTO Michael Freedman talk about the evolution from data lakes (unstructured) and data warehouses (structured) into unified data lakehouses, I had a thought...
May 1, 2025 at 3:14 PM
Listening to our CTO Michael Freedman talk about the evolution from data lakes (unstructured) and data warehouses (structured) into unified data lakehouses, I had a thought...
𝗦𝗜𝗠𝗣𝗟𝗜𝗙𝗬 𝗗𝗢𝗖𝗨𝗠𝗘𝗡𝗧 𝗘𝗠𝗕𝗘𝗗𝗗𝗜𝗡𝗚𝗦 𝗪𝗜𝗧𝗛 𝗣𝗚𝗔𝗜 𝗩𝗘𝗖𝗧𝗢𝗥𝗜𝗭𝗘𝗥: 𝗣𝗼𝘀𝘁𝗴𝗿𝗲𝘀 + 𝗔𝗺𝗮𝘇𝗼𝗻 𝗦𝟯
We've expanded pgai Vectorizer to automatically create searchable vector embeddings in Postgres from documents stored in S3 while keeping the original files in place.
We've expanded pgai Vectorizer to automatically create searchable vector embeddings in Postgres from documents stored in S3 while keeping the original files in place.
April 30, 2025 at 8:13 PM
𝗦𝗜𝗠𝗣𝗟𝗜𝗙𝗬 𝗗𝗢𝗖𝗨𝗠𝗘𝗡𝗧 𝗘𝗠𝗕𝗘𝗗𝗗𝗜𝗡𝗚𝗦 𝗪𝗜𝗧𝗛 𝗣𝗚𝗔𝗜 𝗩𝗘𝗖𝗧𝗢𝗥𝗜𝗭𝗘𝗥: 𝗣𝗼𝘀𝘁𝗴𝗿𝗲𝘀 + 𝗔𝗺𝗮𝘇𝗼𝗻 𝗦𝟯
We've expanded pgai Vectorizer to automatically create searchable vector embeddings in Postgres from documents stored in S3 while keeping the original files in place.
We've expanded pgai Vectorizer to automatically create searchable vector embeddings in Postgres from documents stored in S3 while keeping the original files in place.
PGVECTOR VS QDRANT: You don’t need a specialized vector database for large scale. Postgres is all you need.
April 29, 2025 at 2:59 PM
PGVECTOR VS QDRANT: You don’t need a specialized vector database for large scale. Postgres is all you need.
New database benchmarks for real-time analytics by Timescale.
Blog: tsdb.co/benchmark-rta
Github: github.com/timescale/rt...
Blog: tsdb.co/benchmark-rta
Github: github.com/timescale/rt...
Not All Analytics Are Equal: Benchmarking Databases for Real-Time Analytics Applications
Not all analytics are equal—neither are benchmarks. Meet RTABench and compare different databases—from PostgreSQL to ClickHouse—using real-world query patterns.
www.timescale.com
March 27, 2025 at 12:48 PM
New database benchmarks for real-time analytics by Timescale.
Blog: tsdb.co/benchmark-rta
Github: github.com/timescale/rt...
Blog: tsdb.co/benchmark-rta
Github: github.com/timescale/rt...
We just released a new database analytics benchmark. RTABench is geared to talking about real-time analytic /applications/ (that is how it differs from ClickBench).
March 26, 2025 at 2:47 PM
We just released a new database analytics benchmark. RTABench is geared to talking about real-time analytic /applications/ (that is how it differs from ClickBench).
Our goal: make working with vector data and LLMs as easy as possible with the developer tool you know and love.
🐍 SQLAlchemy + pgai Vectorizer: Seamless Vector Search in Python 🐍
www.timescale.com/blog/pgai-ve...
🐍 SQLAlchemy + pgai Vectorizer: Seamless Vector Search in Python 🐍
www.timescale.com/blog/pgai-ve...
Pgai Vectorizer Meets Python: Integrating SQLAlchemy and Alembic
You can now seamlessly integrate pgai Vectorizer and Python using SQLAlchemy and Alembic to automate embedding creation and management. Learn how.
www.timescale.com
February 20, 2025 at 4:28 PM
Our goal: make working with vector data and LLMs as easy as possible with the developer tool you know and love.
🐍 SQLAlchemy + pgai Vectorizer: Seamless Vector Search in Python 🐍
www.timescale.com/blog/pgai-ve...
🐍 SQLAlchemy + pgai Vectorizer: Seamless Vector Search in Python 🐍
www.timescale.com/blog/pgai-ve...
Reposted by cevian
LGBTQIA minors make up 40% of the endangered minor runaway population. They are 120% more likely to be unhoused. They are also uniquely at risk for exploitation and trafficking.
This is breathtakingly evil.
This is breathtakingly evil.
New — The National Center for Missing & Exploited Children (NCMEC) was told this week by DOJ that they'd lose their funding if the org didn't remove any mentions of LGBTQIA+ issues from their public materials, I've learned. Staff were told they need to deadname trans kids in their reports to comply.
February 7, 2025 at 4:00 AM
LGBTQIA minors make up 40% of the endangered minor runaway population. They are 120% more likely to be unhoused. They are also uniquely at risk for exploitation and trafficking.
This is breathtakingly evil.
This is breathtakingly evil.
Vasily Grossman's "The Road" (the story, somewhat confusingly published in "The Road" collection of essays) may be the single best description of depression I've ever seen in writing. Incredibly, it's told from the vantage point of a Mule!
December 22, 2024 at 8:30 PM
Vasily Grossman's "The Road" (the story, somewhat confusingly published in "The Road" collection of essays) may be the single best description of depression I've ever seen in writing. Incredibly, it's told from the vantage point of a Mule!
and the core pgai repo is here: github.com/timescale/pgai
GitHub - timescale/pgai: A suite of tools to develop RAG, semantic search, and other AI applications more easily with PostgreSQL
A suite of tools to develop RAG, semantic search, and other AI applications more easily with PostgreSQL - timescale/pgai
github.com
December 19, 2024 at 9:17 PM
and the core pgai repo is here: github.com/timescale/pgai
Index types for pgvector dense vector types and their pros and cons.
December 19, 2024 at 3:43 PM
Index types for pgvector dense vector types and their pros and cons.
Testing out a theory about Math education: Can you explain why a times b = b times a (a*b=b*a)?
If so, what's you explanation?
If not, I'd also like to know!
If so, what's you explanation?
If not, I'd also like to know!
December 19, 2024 at 1:47 PM
Testing out a theory about Math education: Can you explain why a times b = b times a (a*b=b*a)?
If so, what's you explanation?
If not, I'd also like to know!
If so, what's you explanation?
If not, I'd also like to know!
📺 How to evaluate different embedding models on your actual data
In this video, Jacky explains a simple (and unsupervised) yet powerful technique to evaluate embedding models using your own data:
www.youtube.com/watch?v=VgSF...
In this video, Jacky explains a simple (and unsupervised) yet powerful technique to evaluate embedding models using your own data:
www.youtube.com/watch?v=VgSF...
OpenAI vs. Open-source Embedding Model Showdown
YouTube video by Timescale
www.youtube.com
December 19, 2024 at 1:34 AM
📺 How to evaluate different embedding models on your actual data
In this video, Jacky explains a simple (and unsupervised) yet powerful technique to evaluate embedding models using your own data:
www.youtube.com/watch?v=VgSF...
In this video, Jacky explains a simple (and unsupervised) yet powerful technique to evaluate embedding models using your own data:
www.youtube.com/watch?v=VgSF...
State of PostgreSQL results for 2024 are now live!
Find out what's been going on in the community and ecosystem.
www.timescale.com/blog/state-o...
Find out what's been going on in the community and ecosystem.
www.timescale.com/blog/state-o...
A Sneak Peek Into the State of PostgreSQL 2024
The 2024 State of PostgreSQL survey report is out! Read its first findings.
www.timescale.com
December 17, 2024 at 9:53 PM
State of PostgreSQL results for 2024 are now live!
Find out what's been going on in the community and ecosystem.
www.timescale.com/blog/state-o...
Find out what's been going on in the community and ecosystem.
www.timescale.com/blog/state-o...
We’ve added support for Voyage AI in pgai Vectorizer, our open-source tool for auto-creating and syncing embeddings in PostgreSQL.
December 17, 2024 at 6:33 PM
We’ve added support for Voyage AI in pgai Vectorizer, our open-source tool for auto-creating and syncing embeddings in PostgreSQL.
AI doom scenarios often feel a bit off to me—not because they're wrong about AI being risky, but because they imagine the danger as some AGI-built Terminator rising up to enslave humanity. Honestly, the real risks seem a lot more... human.
December 16, 2024 at 6:32 PM
AI doom scenarios often feel a bit off to me—not because they're wrong about AI being risky, but because they imagine the danger as some AGI-built Terminator rising up to enslave humanity. Honestly, the real risks seem a lot more... human.