Executive Summary

A leading European Original Equipment Manufacturer (OEM) specializing in Baggage and Cargo Handling sought the expertise of Newtuple Technologies. Their goal was to enhance the accessibility and ease of data consumption for airport support staff through the integration of a user-friendly chatbot conversational interface over their existing APIs of the data platform.

The baggage and cargo data resides in a complex network of databases, traditionally necessitating specialists to navigate multiple APIs for information retrieval. Newtuple Technologies created a GenAI based chatbot solution layer on top of these APIs which accepts natural language prompts and simplifies this process. The final solution can be accessed using a web based application.

X%

Increase in 

Y%

Time saved in running data queries

Z%

Increase in adoption of data platform

Problem Statement

The data storage system is segmented into approximately five distinct verticals, each categorized by the type of data being stored. These verticals each possess their unique set of APIs for data access. This structure necessitates users to often engage multiple APIs based on their specific query and combine the data to arrive at the final response to the query. This process, which requires knowledge of calling multiple APIs is inherently complex, poses a significant challenge for the average user to efficiently learn and navigate the system.

Our solution solves this challenge in two steps. First we interpret the natural language queries and translate them into a plan that the AI generates, which consists of a sequence of steps consisting of API calls, data joins, transformations and inferencing on data. Second, it executes these steps in the backend step by step. Finally the final solution is displayed to the web based UI. This innovative approach democratizes access to the data platform, enabling users of diverse backgrounds to effortlessly extract the information they need.

The entire process of creating our solution, from the initial concept to its final deployment, was completed in a span of eight weeks. Throughout this period, we employed an agile methodology, which allowed us to effectively integrate user feedback directly into the development of the solution. Our clients saw a prototype of the application within 1 week of kick-off, enabling them to immediately provide the team with actionable feedback.

High level objectives

• Create user friendly access to data through web application

• Drive up adoption of data platform through natural language queries

• Reduce time taken for executing queries

The solution involves the following components:

1. User interacts with Web based UI designed in Chainlit which offers a chat interface for providing queries.

2. Task Planner uses an LLM to create a step-by-step plan of tasks to perform based on the user query and API specifications. The past responses of the planner are stored in a Knowledge Base for increasing the planner performance.

3. The steps prepared by the planner are executed by Task Executor engine which can involve either one of the following

   • Generator engine which calls the APIs using API Executor post validating the schema using API Validator

   • Transformer engine which performs data wrangling. In few instances it can convert the API responses into pandas DataFrame objects and extract information using DataFrame Handler

   • Summarizer engine which is used to summarize the output generated by Generator and/or Transformer in human readable format to be sent back to UI

   • Clarifier engine which is invoked in cases where the user prompt is missing some key information and needs clarification. After clarification, the Task Planner engine uses the prompt to proceed with normal execution as explained in steps above