Thanks to @nicolay.fyi for giving me the opportunity to talk about @trustgraph.bsky.social on How AI Is Built! Labels such as "facts", "observations", and "assertions" take on new meanings when we begin to consider time. Click 👇 to watch the full episode! 🎙️ youtu.be/VpFVAE3L1nk?

Dropping some new episodes on @howaiisbuilt.fm . Links below.

Trey and I talk about the different techniques for AI-powered search and how we can combine them to build modern search systems. Spotify: open.spotify.com/episode/1udV... Apple: podcasts.apple.com/us/podcast/a...

You want the exact opposite. You want layers of tools aligned in a graph that you can tune, debug, and update in isolation. Today on How AI Is Built, we are talking to one of the OGs of search: Trey Grainger, the author of AI Powered Search. www.youtube.com/watch?v=6IQq...

New episode is out. The three contexts of search, layered architectures and much more!

Youtube: www.youtube.com/watch?v=trG5... Spotify: open.spotify.com/episode/6eyT... Apple: podcasts.apple.com/us/podcast/c...

The biggest lie in RAG is that semantic search is simple. The reality is that it's easy to build, it's easy to get up and running, but it's really hard to get right. And if you don't have a good setup, it's near impossible to debug. One of the reasons it's really hard is chunking.

Most companies can't afford huge teams labeling AI data. So, use an AI model to train an AI model. The big labs like Cohere and OpenAI already use “synthetic data” - AI-generated data that mimics real-world patterns. The LLMs you use are already trained with it. youtu.be/thqgKG5lZ8Q

Youtube: youtu.be/Z9Z820HadIA Spotify: open.spotify.com/episode/3LAJ... Apple: podcasts.apple.com/us/podcast/a...

Want to learn more? Today on @howaiisbuilt.fm, we are talking to Stephen Batifol from Zilliz. Stephen and I discuss agentic RAG and the future of search - where systems decide their own path to find answers. What's your take on agentic RAG? youtu.be/Z9Z820HadIA

"Instead of being a one-way pipeline, agentic RAG allows you to check, 'Am I actually answering the user's question?'" Different questions need different approaches. ➡️ 𝗤𝘂𝗲𝗿𝘆-𝗕𝗮𝘀𝗲𝗱 𝗙𝗹𝗲𝘅𝗶𝗯𝗶𝗹𝗶𝘁𝘆: - Structured data? Use SQL - Context-rich query? Use vector search - Date-specific? Apply filters first

We talk about how they enable BM25 on PostgreSQL, how they integrate into the Postgres Query engines, and how you can build search in your database. open.spotify.com/episode/4CXX...

Not anymore. ParadeDB is building an open-source PostgreSQL extension to enable search within your database. Today on How AI Is Built, I am talking to @philippemnoel.bsky.social , the founder and CEO of @paradedb.bsky.social. youtu.be/RPjGuOcrTsQ

Many companies use ElasticSearch or OpenSearch and use 10% of the capacity. On top, they have to build ETL pipelines. Get data normalized. Worry about race conditions. All in all, when you want to do search on top of your existing database, you are forced to build distributed systems. #ai

Documentation quality is the silent killer of RAG systems. A single ambiguous sentence might corrupt an entire set of responses. But the hardest part isn't fixing errors - it's finding them. Check out the episode with Max. Links to Spotify, Apple in the thread.

LLMs hallucinate. We want to put the blame on them. But often it’s our fault. Many knowledge bases have: → Temporal Inconsistencies - Multiple versions from different time periods - Historical information without timeline context >>

With RAG these issues are amplified. We do not look at full documents anymore, but at bits and pieces. So we have to be extra careful. Today on @howaiisbuilt.fm we talk to Max Buckley. Max works at Google and has built a lot of interesting stuff with LLMs to improve knowledge bases for RAG. >>

Some query types might not work at all. It is very costly in terms of storage and compute. We have to keep our indexes in memory to achieve a low enough latency for search. What we are talking about today works for everything, works out of domain, and is one of the most efficient. >>

People implementing RAG jump straight into vector search. But vector search has a lot of downsides. Vector search is not robust out of domain. Different types of queries need different embedding models with different vector indices. >>

You probably guessed it, we are talking about the OG ranking function in search: BM25. Today we are back continuing our series on search on @howaiisbuilt.fm with @taidesu.bsky.social. We talk about BM25, how it works, what makes it great and how you can tailor it to your use-case.

"Sadly, it's a bit off a snake oil. These long context embedding models have tested basically all of them, not really working well. So it's [best length of chunks] something between like 500 and 1,000 tokens." Text embeddings are far from perfect. They struggle with long documents. >>

Vector Databases come with their own set of challenges. The data is too large to be stored on a single node. We often need to handle 10k to 50k QPS. Indexes are very slow to build, but we still want to search the fresh data. >>

Catch the episode on: - Youtube: youtu.be/3PEARAf7HEc (now in 4K :D) - Spotify: open.spotify.com/episode/5lCl... - Apple: podcasts.apple.com/us/podcast/v...

“There is no free lunch.” Every performance optimization comes with tradeoffs in either functionality, flexibility, or cost. When building search systems, there's a seductive idea that we can optimize everything: fast results, high relevancy, and low costs. But that’s not the reality.