Lightnews — Scholar-powered news

Philippe Noël @philippemnoel.bsky.social · 3d

All good, I think we're still a step before doing that

Philippe Noël @philippemnoel.bsky.social · 4d

I'm interested in hiring an "SEO consultant" to come in, evaluate our SEO posture and make a list of recommendations for what to improve next. Got anyone to recommend?

1 1 2

Reposted by Philippe Noël

Chris @chris.blue · 9d

Had a few folks ask why you'd need ACID for search. This @paradedb.bsky.social article from @philippemnoel.bsky.social and company does a decent job making the case.

There are a lot of cases (e.g. fintech) where strong consistency and durability is a big deal for search.

ParadeDB

The Transactional Elasticsearch Alternative Built on Postgres

www.paradedb.com

1 2 10

Philippe Noël @philippemnoel.bsky.social · 11d

no clue

Philippe Noël @philippemnoel.bsky.social · 13d

There's a ParadeDB pg_search package for the .NET! Kudos to Nandor on this one. Source and package manager links:

- www.nuget.org/packages/EFC...
- github.com/nandor23/EFC...

EFCore.ParadeDB.PgSearch 0.1.0

EFCore.ParadeDB.PgSearch adds ParadeDB's pg_search extension support to Npgsql.EntityFrameworkCore.PostgreSQL. Exposes ParadeDB search functions through EF.Functions API for LINQ-based full-text searc...

www.nuget.org

3

Philippe Noël @philippemnoel.bsky.social · 16d

I just try not to run too many apps at the same time now :(

1

Philippe Noël @philippemnoel.bsky.social · 18d

So annoying. Need the next update to fix this haha

1

Philippe Noël @philippemnoel.bsky.social · 18d

An excellent, to-the-point blog post. One architecture difference to note is that MongoDB Search runs outside of the main MongoDB process, while in ParadeDB it runs inside.

We've revamping our tokenizers and will fix this emoji tokenization issue. Thanks Franck!

Franck Pachot @franckpachot.bsky.social · 20d

Comparing 🍏 to 🍊, and MongoDB to PostgreSQL, with full-text search indexes

dev.to/franckpachot...

Text Search with MongoDB and PostgreSQL

MongoDB Search Indexes provide full‑text search capabilities directly within MongoDB, allowing...

dev.to

4

Philippe Noël @philippemnoel.bsky.social · 18d

Is it just me or the new macOS Liquid GLass Spotlight is kinda laggy... Ugh (On M3 MacBook Air)

1 3

Philippe Noël @philippemnoel.bsky.social · 19d

A fantastic podcast by our engineer, Ankit Mittal, on how Instacart built search on Postgres, where the future of search infrastructure is going, and why that led him to join us at ParadeDB: www.youtube.com/watch?v=hohp...

Postgres vs. Elasticsearch: Instacart’s Unexpected Winner in High-Stakes Search with Ankit Mittal

YouTube video by The Data Engineering Show - Podcast

www.youtube.com

2

Philippe Noël @philippemnoel.bsky.social · 20d

What is your take on Elasticsearch shipping a SQL interface?

1

Philippe Noël @philippemnoel.bsky.social · 20d

I'm not as familiar with this use case, but I don't see how this couldn't be declared as SQL.

1

Philippe Noël @philippemnoel.bsky.social · 20d

We came up with this categorization internally:

- real-time search (e.g. dashboards, SaaS search boxes)
- observability
- RAG/GenAI

Philippe Noël @philippemnoel.bsky.social · 20d

Instacart can know this because they have recommender systems trained on what this user might want.

But I don't think all search use cases need this. I suspect AI-search use cases will to a lesser extent, but that it'll be beneficial for them. That's more likely to be an area we optimize around.

1

Philippe Noël @philippemnoel.bsky.social · 20d

By arbitrary, you mean custom code / ML models? I think these matter a lot in a use case like e-commerce where a lot of domain knowledge and user-specific knowledge is applied. For example, an Instacart user might type "banana" but they want "banana yogurt" rather than "bananas".

2 1

Philippe Noël @philippemnoel.bsky.social · 20d

I think it's important to define the use cases well. ParadeDB is not focusing on e-commerce, and so we have a lower emphasis on re-ranking. Elastic has good capabilities there since they are so broad they target pretty much all search use cases.

1

Philippe Noël @philippemnoel.bsky.social · 20d

Blog post written by Ankit (who now works at ParadeDB) here: tech.instacart.com/how-instacar...

How Instacart Built a Modern Search Infrastructure on Postgres

Ankit Mittal, Vinesh Gudla, Guanghua Shu, Akshay Nair, Tejaswi Tenneti, Andrew Tanner

tech.instacart.com

2 2

Philippe Noël @philippemnoel.bsky.social · 20d

Thhey're very powerful, especially in e-commerce use cases. One of our engineers used to work on Instacart's search team. Their use case fits exactly what you describe -- Heavy emphasis on re-ranking and ML models and somewhat limited emphasis on keyword relevance. They're built on Postgres and SQL

1

Philippe Noël @philippemnoel.bsky.social · 20d

ParadeDB has somewhat limited re-ranking capabilities today, but that hasn't stopped our customers from seeing value and I expect we'll be improving our re-ranking capabilities heavily in 2026.

1 3

Philippe Noël @philippemnoel.bsky.social · 20d

I agree that re-ranking is important once you enter use cases where result relevance at the margins is critical. For example, Best Buy might be interested in a marginal relevance improvement given the sheer volume of sales they process.

But for many, a simple Top K and fast sort/ordering is enough

1 1

Philippe Noël @philippemnoel.bsky.social · 20d

I'm with you until you mention that "this is beyond what SQL can provide". You're saying SQL can't provide re-ranking capabilities? I'm not sure I follow why that'd be the case.

1

Philippe Noël @philippemnoel.bsky.social · 20d

ParadeDB uses are similar to Elasticsearch use cases. The most common is something like "Count(*) + Top K". E.g. From a dataset of say 500M documents, return all that match "keyboard" by relevance, sort by ranking, and return the top 50 so I can paginate"

1

Philippe Noël @philippemnoel.bsky.social · 20d

Yes. Many industries require read-after-write guarantees on their search, since it's how their users interact with the data and correctness is critical (e.g. searching legal text, or financial transactions)

You also get JOIN support, simpler infra, lower downtime, etc.

1 4

Philippe Noël @philippemnoel.bsky.social · 22d

We're writing again! We've got a series of exciting articles coming up, with the first one: Elasticsearch Was Never a Database

Elasticsearch is an amazing product. Really. But it's not a transactional database. Yet because of how painful sync-ing it with Postgres is, many try to. Read on:

ParadeDB @paradedb.bsky.social · 22d

People who use Elasticsearch fall into two camps: those who know it shouldn’t be used as a source of truth, and those who haven’t thought it through…

Seriously though, if you use Elastic as a database, we’d love to hear about it.

www.paradedb.com/blog/elastic...

#elasticsearch #database #search

ParadeDB

The Transactional Elasticsearch Alternative

www.paradedb.com

3 1 13

Philippe Noël @philippemnoel.bsky.social · Aug 29

Interesting article on narrowing the use case to find PMF: medium.com/craft-ventur...

The Sharp Startup: When PayPal Found Product-Market Fit

Reflections on the 20th anniversary of a $100+ billion product

medium.com

1 2