Vigil

Slack's Notification Flowchart
Temperature vs Top-K

Inspiration

What it does

Vigil elevates issues related to your current focus as they arise, saving you from manually scanning threaded communications.

How we built it

We use LlamaIndex's Typescript library to vectorize and store Slack messages in Datastax' AstraDB, which we query in conjunction with a user's Slack status to find related messages.

Challenges we ran into

We struggled to find queries that did not return nodes, especially irrelevant ones. Spent maybe too much time trying to pass a NodePostprocessor to LlamaIndex' VectorStoreIndex#asRetriever method expecting it to behave like VectorStoreIndex#asQueryEngine... only to switch to #asQueryEngine in the last 45 minutes before the deadline!

Accomplishments that we're proud of

Neither of us had written to, or read from, a vectordb before this project, and I hadn't previously vectorized embeddings at all! So going from 0-to-1 in RAG patterns, able to augment queries with arbitrary documents, is a real thrill.

Also came up with a lovely little diagram to illustrate the relationship between temperature and top K that I think is pretty cute.