How we built an NLP Based Search Engine

AI-powered BI tools are transforming the business industry worldwide. They are helping businesses by:

  • Providing them with day-to-day insights into the changing business world, thus, helping them make informed decisions.
  • Helping them comb through heaps of unstructured data and making them insight-rich instead of just data-rich.
  • Introducing data democratization in their businesses and removing any gatekeepers creating bottlenecks at the gateway to the data

We helped a client do the same.

The Client:

The Client provided AI-based Business analytics to investors to help them make better decisions.

The Problem:

The client wanted to automate the process of finding competitors for a given company. The problem was – There were 10 million companies to search across and numerous other features to consider per company, apart from the company description. Moreover, a real-time solution was needed. Batch-processing wouldn’t simply work, since the features were updated frequently and the algorithm could be influenced by the end-users’ decisions on the UI.

The Solution:

We came up with an AI-enabled solution to rank competitor companies’ reliably. Text fields were analyzed and compared using NLP algorithms and inputs from other fields were subsequently fed into the algorithm to refine the ranks.

At the core, our solution used the following workflow:

  1. We first used the BM25 algorithm to crudely rank descriptions.
  2. Then, a RoBERTa model fine-tuned on the specific task of finding competitors from a proprietary dataset was used to generate embeddings.
  3. Finally, we used an efficient embedding search framework developed internally at Aidetic to enable superfast cosine similarity calculation.

Results:

By using Aidetic’s custom search engine for ranking similar documents, the client was able to serve this information in real-time on their dashboard, with a response time of less than 1 second. We helped develop the backend framework but also helped deploy the services on the cloud in a way that was optimized for availability, faster response time, and cost-effectiveness.

So, though in its early stages, AI for BI is the future as predicted by the world technology leaders. If you’re looking to leverage the enormous potential of AI for your business, give us a call.

Technology Stack:

Leave a Reply