Links of the Presentation: Experimenting with AI and LLM to make docs searchable through a chat application
These are all the links of my presentation at the Fosdem Conference in Brussels, Saturday, February 3th, 2024, 17:40 CET.
Experimenting with AI and LLM to make docs searchable through a chat application
ChatGPT is considered to bring the end to search engines. But it’s also a pretty good liar, just making stories up. As chat-based search engines, powered by machine learning and artificial intelligence, use an LLM (Large Language Model), they have no other knowledge than an algorithm that predicts the most likely word that should follow the previous ones. On top of that, these LLMs are not up-to-date, as they are generated with a “snapshot” of available data at a specific date when the LLM is being developed.
But how do we cope with the challenge that our managers and end-users want us to provide a chat-based user interface that produces correct results?
At Azul, all our public knowledge about Java and the products we deliver is combined on docs.azul.com. It’s a “classic documentation” website, generated as a static website based on AsciiDoc files. As Java lovers, we experimented with several free, open-source tools to make our docs chat-searchable. Join this session to learn if we can dump traditional docs and enter the chat age…
About Me, Foojay, and Azul
- Mastodon: foojay.social/@frankdelporte
- Personal blog
- Foojay author
- Foojay podcasts
- Working at Azul
- Azul Documentation
Books
Links
- GPT-3 on Wikipedia
- GPT-5 expectations
- Phind: chat with links to sources
- Your documentation is used for training
- Vaadin chat with documents
- Pinecone: embeddings vector database
- OpenAI Chat API
- Lize Raes at Devoxx Belgium, demonstrating LangChain4j
- My demo projects:
- Java and LLMs: are we there yet?
- jan.ai: Bringing AI to your Desktop
- What can possibly go wrong?