This repo is behind a small website where you can find events in FOSDEM 2024 based on content search or via connections.
I wrote this because, whilst I really enjoy attending FOSDEM, the amount of possible things to see is large and there is always a chance I can miss something.
The main website does have search, so, to be totally honest I should also say I just wanted to play about with OpenAI Embeddings. However, one outcome is that you'll find results for queries here you won't find on main site e.g. "controversial"
Indexing:
-
Event Content is extracted from the Pentabarf version of the Schedule
-
The OpenAI Embeddings are looked-up for the title, abstract, and any other relevant content in the Event info we can derive from the Schedule. Additionally:
- if slides are available, these are downloaded, text is extracted using Apache Tika Server, and these are added to the input to the embedding
-
Events and Embeddings info are saved in JSON files; these are the
index
Lookup:
- fly.io service used to host website and handle queries
- JSON index files are loaded into memory on startup
- For each query:
- Search query is converted live to an OpenAI Embedding
- Nearest match found via vector distance search, using
nalgebrafor distance calculation