Introducing Contextual Answers

AI21 has unveiled Contextual Answers. It is an end-to-end API designed so that when an enterprise uploads a document library such as a knowledge base, help center, reports, policies, guidelines, manuals, and playbooks, the answer engine responds based only on that corpus. The goal is to reduce the cost and complexity of generative AI that isn't tailored to an organization's data, as well as hallucination.

AI21 also cited the Clarivate case, explaining that it can be used in document-search-heavy tasks like education, customer support, finance, legal, insurance, sales, and field technical support. From the perspective of internal knowledge management systems (KMS), a Coveo report found that employees spend an average of 3.6 hours a day searching for relevant information.

The implementation flow is simple.

• Step 1: Upload files to the Library, which offers free storage, via Studio, the Python SDK, or HTTP requests.
• Step 2: When a user enters a question, the retriever finds relevant context across the entire knowledge base, and the model returns an answer along with its source references.
• Step 3: If the answer isn't in the documents, it returns an empty response to avoid speculative answers.

For example, it returns 2 days a week for a question about remote work days, while for questions not covered in the documents, such as meal allowances, it provides no answer. It also supports grouping documents by directory or label to query only a subset of them.