Semilab

Hiflylabs created a Microsoft Azure pipeline that enables OpenAI’s model to be fed corporate information and answer user questions in real time—while complying with all corporate safety and privacy standards. Existing data preprocessing techniques were updated—improving their efficiency by 10% on average. 

With a new chunking strategy, we got more coherent content blocks and better text recognition. Retrieval augmented generation was implemented, improving searchability. An automated evaluation system was developed as well to save time on iterative cycles. 

The successful proof-of-concept initially turned into a multi-sprint project with the intent of developing a customized model with full access and integration into the company data infrastructure.