NLSQL use case sample based on the USA transportation data
Slack text interface to census gov data
Natural Language User Interface in Slack
Text to SQL API for Natural Language Interface in Slack
Slack interface to database usage
TOP products, TOP countries queries in Slack interface to database
picture data visualisation in Slack
In a few years, Artificial Intelligence will be almost intelligent and understand everything in your database. It would understand not just what you wanted, but find the right pieces of data, do the necessary calculations, and give you the answer you need. What is the latest revenue forecast for Q3 2022 in the UK market? No problem. A few seconds, and it’ll spit out the information you need. Faster and more reliable than analysts.
While we’re not there yet in NLSQL, we are getting closer, and what we have right now is pretty powerful if we say so ourselves.
Picture this: you’re talking numbers with someone else during business negotiations. They throw out a figure and it can change everything about the negotiation. You could quickly ping an analyst and ask them for the necessary figures, however, it might take them half an hour to query, compile and analyze the data before they can send it your way. Or you could ping NLSQL, get the price, the turnover, sales, inventory, and run the figures yourself. You can be completely confident in the numbers at your fingertips and come to a negotiation result that’s a win for everybody.
What it does
NLSQL is a Slack Application that is accessible only for company employees. It can support users not only with prebuilt dashboards but with custom queries and data visualizations based on natural language questions asked.
For Digital Hackathon we developed free of charge Slack Application, so anyone can ask for export-import trade balance transactions data from the USA 🇺🇸 worldwide 🌎 based on census gov open data. Anyone can add it into Slack and test how it works by asking questions about the USA import for different countries and product groups.
How we built it
Open-source Machine Learning Algorithms such as Seq2SQL, TREQS, PtrGen are not accurate and reliable enough in order to be used in the production environment, so we developed our own NLSQL AI engine using a new programming pattern to improve significantly upon these solutions. We are unique in 4 main areas, where we excel from the current state-of-the-art: speed, accuracy, ease-of-integration, and user experience.
One of the most significant improvements in the accuracy of results. NLSQL reached 98% accuracy compared with Seq2SQL 9%, TREQS 56%, PtrGen 18% within the same MIMIC-III dataset from MIT. Moreover, mentioned models require database structure modifications, while NLSQL works with the existing database structure and doesn't require any database changes.
NLSQL software combines functional and imperative programming concepts and AI planning patterns in a special way, that is our Trade Secret. However, we are aware of the prior art, and our IP is protected by UK laws and the Intellectual Property Office.
As complementary software: we used is Plotly/MatplotLib for great data visualizations, MS Bot Framework, and Azure Bot for fast integration into Slack and other messaging platforms. We used cloud-native software architecture for the ability for horizontal scale-up and quick vnet integrations.
Challenges we ran into
The biggest challenge we ran into is scalability because of different database structures for different companies. But we've managed to create a fully self-service customization process on our NLSQL platform. Anyone can easily emulate their database structure on our platform and generate API token following our video tutorials.
Also during our cust dev interviews, we were informed that usually, corporate data security doesn't allow to access their data for third parties. For solving this challenge, we were able to modify our software architecture, which doesn't require any sensitive or confidential data access or transfer outside the corporate IT ecosystem. NLSQL bot system can work on customer cloud with NLP to SQL API, that translates Natural Language to Structured Language Code. Cloud deployment provides us the ability for real-time Natural Language Understanding Improvement and the best end-user experience ever.
Database structures and terminology can be complex for intents detection, so we've created a model that can ask additional questions to end-user if something is not clear enough from the entire question. Asking additional questions in a conversation mode is another of our competitive advantage compared to similar tools or software on the market.
Accomplishments that we're proud of
We've won the SAP SMB Hackathon with a use case based on the SAP Business One sales data using the SAP B1 standard API layer last year. Data and test SAP B1 system were supplied by SAP within that Hackathon. As there is no data sample, we've used open import-export trade balance transactions data for Digital Hackathon this year. We are proud to contribute our use case with import-export trade balance data from the census gov as part of the Digital Hackathon.
NLSQL software is officially certified by SAP for the most innovative product SAP HANA with the world's best in-memory database. All security vulnerabilities tests of the SAP ICC team are successfully passed by NLSQL. NLSQL is an SAP Partner with a listing on SAP App Store.
We've raised a German Federal Ministry funding as the most accurate software for Healthcare BI development for German Healthcare system needs.
What we learned
Natural Language Processing is still a challenging space, but with AI planning concepts and the specific focused task on transformation from Natural Language to Structured Query Code with emulated database structure, it can provide accurate calculated responses.
Slack is a perfect End-User interface for our application with great API and resources to provide company employees with our application in a secure and easy-to-use manner.
What's next for Chat Interface To Complex Corporate Databases By NLSQL
Next, we want to create a business plan for raising investments for further product development and commercialization
We plan to create the best ever product with the highest accuracy and end-user experience. Complex B2B Software integration takes time, but we are going to automate the setup and customization process with Azure automation tools, so anyone will be able to install NLSQL using Azure Marketplace in a couple of minutes.
NLSQL is part of Microsoft for startups program, so we have access to their knowledge base and Azure software architects for smooth and secure integration into popular databases such as Snowflake, Redshift, Microsoft SQL, MySQL, PostgreSQL, Oracle, Cassandra, MongoDB, SAP HANA, etc.