Entity linking is a cute problem in NLP where your are tasked with connecting a mention to an entry in a knowledge base. For example, take the sentence "In order to save Troy, Paris had to be sacrificed" -- the mention Paris should connect to Paris (mythological figure) not the capital of France.
I tackle this problem by training bi-encoders and embedding mentions and entities to the same space.
I also wrote a datastructure in C++ that uses mmap to store and retrieve structural data from a knowledge base.
Here is an experimental repo showing that LLM reranking is harder than it seems.
App that supports learning using active recall questions.
Small utility to track what happend during the day.
Here, I gather small AI projects or experiemnts.
Recently, I've became interested in constrained programming. Here are solutions to some practice problems in Picat language.
Implementation of common decision tree algorithms in Julia.