May 2022 Summaries

A question answering model requires data to learn how to extract answers from it, as the models themselves do not contain any answers. The data used to train a model is referred to as context, which the model uses to search for and extract relevant information from. Pre-trained general-purpose models can often be sufficient for many use cases, but fine-tuning with domain-specific data may be necessary in certain scenarios. The process of providing data to a question answering model involves storing it in a database or data storage solution that can efficiently retrieve and pass on small pieces of text to the model, one after the other, while also handling the complexity of formatting and efficiency requirements. Solutions such as Haystack provide functionality for this journey from data to answers, allowing developers to define pipelines that combine multiple components, including databases, retrievers, readers, and more, to efficiently provide data to a question answering model.

At deepset, the interview process is designed to be transparent and informative, with a focus on finding candidates who not only match the company's values but also share its mindset. The process consists of two preliminary interviews, followed by three steps: a practical challenge, an in-depth interview, and cultural fit interviews. Each step aims to assess the candidate's skills, approach to problem-solving, and cultural fit with the team. The company emphasizes communication, teamwork, and humility, and encourages candidates to ask questions throughout the process. With an 80% offer-acceptance rate, deepset has refined its interview process to ensure a smooth experience for both candidates and the company.