Intelligent Neural Search

Intelligent Neural Search Chatbot

Using this AI-enabled chatbot interface users can retrieve information and documents from a document repository in response to user-provided queries. Employers can use the chatbot to onboard new employees, and students can use it to quickly locate information in study materials documents. Another question-generating module has been developed to provide an instant answer without consuming many AI resources during a conversation.

  • Large number of documents make searching for particular information difficult and time-consuming
  • Question answering in natural language is still not as straightforward as we want. Most of the models are constrained by the training data format
  • Since a query can point to multiple similar documents to retrieve the correct information, the search space needs to be restrained properly
  • Since the neural search has been deployed using a chatbot, it becomes important to preserve users’ context and relevant information from previous queries
  • Managing document access to different user roles and organizations
  • Effective storage of document embeddings for fast response retrieval

The AI-ML solution is made of two levels. In the first layer, questions and answer pairs are generated and stored in a database for each new document that is deposited into the repository. The second layer of the solution is more detailed. This layer enables the extraction of relevant portions of a document in addition to the retrieval of information.

Developed and deployed the chatbot as an extension. Also, built an admin panel as part of the product. The admin panel is a web application that helps in user management and data management- user can upload their data and start the pre-processing pipeline from here. Every uploaded document is stored in an embedded vector form. Another question-generating module was prepared, which once executed, creates question-answer pairs from the documents and stores it in the database. This pre-generated data is used to retrieve responses much faster. A much more complex model is also built, which can provide long answers for a given query.

  • Automate information retrieval process in a more precise Q&A format.
  • Increased customer engagement in both B2B and B2C segments.
  • Helps in reducing time and effort of users, by many folds