Skip to main content

14 posts tagged with "AI"

View All Tags

· 11 min read
Christoph Beck portrait

We are currently implementing semantic search for the 18.0.0 release of OpenProject, which will go live in December. The incredibly skilled AI team at OpenProject, which we collaborate with, has thought about almost every scenario that could arise when an existing product starts to use semantic search.

But this is not the norm. Many clients we have worked with have failed with their first, seemingly simple, transition to semantic / combined search.

So here it is: The "best of" of mistakes you can make when setting up semantic search!

· 12 min read
Christoph Beck portrait
The dev team receives the envelope with the application specs from a dead letter box in the park at night. Two months later, they deliver the code. Their client deploys their app, and everything works as expected. So what was in that envelope?

· 22 min read
Christoph Beck portrait
Agathe Lenclen portrait
Legacy software is the code we lost confidence in. We have build Surveyor, a tool that lets you extract specifications from legacy software. See how Surveyor allows you to reclaim ownership of legacy projects

· 13 min read
Christoph Beck portrait
Joel Koch portrait
With logits processing, you can force open source LLM to comply with a schema you provide. Grammar Constrained Decoding saves you the process of resubmitting malformed LLM responses. Here we show the process with Bumblebee, and explain the background. Want a specific JSON format reliably? Read on!

· 10 min read
Christoph Beck portrait
Joel Koch portrait
Ever tried to get a poem from an LLM doesn't contain the letter `e`? Large Language Models are kind of amazing and surprisingly unreliable at the same time. Using Elixir's Bumblebee and open source LLMs, you can get much better control over the generation. In this article, we introduce you to logits processing, and how it can be used to achieve what ChatGPT can not accomplish. Find out how!

· 8 min read
Christoph Beck portrait
Your search works… but is it actually good? In our previous post, we showed why full-text search struggles with the long, natural-language queries common in MCP interaction. In this follow-up, we dive into dataset generation, evaluation, and search improvement. How do you know if a new embedding model is really better?"