Make Database Queries With Real-Time Chat

We had the opportunity to meet with Adrien Schmidt, CEO and Co-founder of Bouquet.ai during the IT Press Tour in San Francisco. Adrien started building Aristotle, with Natural Language Processing (NLP) for analytics, in 2017, to make data and insights available for people on the frontline of an organization given their need for speed and data at the edge of the organization (e.g. a CSR on the phone with a customer or an airline ticket agent at the airport with a passenger).

People are the edge want an answer to their question — a number, a growth rate, not a graph. You need the answer to your question to make data worthwhile. Adrien does not want users to have to interpret results since they just want an answer to their question.

Bouquet.ai helps the enterprise automate the creation of NLP models in the context of analytics, on proprietary databases, in dynamic environments, at scale. Target users are business experts, not predictive modeling experts that can execute the process in a dynamic environment.

To automate the creation of NLP models to voice activate a database, they start by identifying the problem to solve with the thought of where does this answer lead. The objective is to provide the user with the answer, not a link to click and read the answer. The platform is mobile-first since mobile devices are frequently what the end-user is using to query the database.

The platform is designed to recognize dialog, not search with the thought being if you are going to voice activate a system you are not going to voice out words. You want something interactive that understands what you are asking in a natural language dialog. They achieve disambiguation around names by embedding search. The initial glossary is a set of synonyms that is a key feature of the product. User-defined terms are learned. Bouquet.ai supports the most commonly asked questions in the common use case of an end user

Aristotle is a public app on Slack. Search is based on Elastic Search and the backend is a dynamic SQL Query with joins and access to date ranges. They use a command-line interface (CLI) to do the modeling. Every time you make modifications, you need to train a new model and that needs to be able to take place quickly. There is an API connection to the client database and Bouquet.ai.

The biggest challenge to adoption is that people are not used to chatbots to access information. However, more than 50% of searches will be voice by 2020.

Brick-and-mortar retail is the vertical where they have seen the most interest. The capacity to quickly find and search is gaining traction. They are also seeing growth and adoption in finance around trading and real-time information.

According to Adrien, the future includes collaborating with BI tools to add natural language queries to their stack. Since the platform is API-based, it can integrate with other chatbots to be able to talk to your databases. He foresees integrating into operational systems like Salesforce, ERP, and marketing automation to enable queries of all these platforms on the fly.