Overview
The Semantic Router is an open-source text classification model that serves as a superfast decision-making layer for LLMs and agents. This model utilizes the power of semantic vector space to expedite decision-making processes, routing requests based on semantic meaning rather than relying on slower LLM generations. The project features a hybrid approach, combining both sparse and dense vectors for text classification.
Semantic Router video demo by James Briggs - Semantic Router Founder
My Contribution
I developed a TF-IDF encoder and integrated it into the hybrid layer as an additional feature. This was the second sparse encoder incorporated, requiring further modifications to the hybrid layer to accommodate the inclusion of new documents. To maintain the program's efficiency, I constructed the algorithm independently, without relying on an existing NLP framework. Furthermore, I created unit tests to ensure the reliability and maintainability of the code throughout the project's progression.
Key Skills
Python, collaboration, NLP, tfidf encoder, algorithms, pytest, git, LLMs, sparse and dense vector integration, model optimization, independent algorithm development