Sigmoidal Success Story




Time to response


Documents processed


Go-Live date

What was the business problem?

With the multitude of medical-related websites and portals, the quality and safety vary widely, and some users are reluctant to access mainstream sites. People are vulnerable to misinformation and exploitation: there is evidence that online medical-related unscientific articles have caused severe health issues among the population. There are solutions designated to counter the dissemination of health information on the internet. For that reason, various strategies are being developed to assess and screen site quality. 

Our client, a highly innovative medical start-up, wanted to develop a platform, providing physicians, nurses, practitioners, and general personnel of hospitals with precise and validated medical data.

What savings and profits does the client achieve?

Sigmoidal delivered a market-ready platform designated for medical-related professionals, providing them with speedy access to information. Our solution offered a unique and cost-effective discovery package for medical libraries, increasing our clients’ sales by roughly 3.2 mln in Q1 2021 (compared with Q1 2020). Sigmoidals’ platform puts authoritative and reliable scientific and health content at the disposal of professionals, effectively helping them be more informed and effective, with 91% answering specialized medical queries on average.

How did we accomplish it?

Sigmoidal came up with a very adaptable design, of a platform, with a live HTTP engine, that leveraged Natural language processing solutions. That was able to answer a wide range of medical queries. Instead of having a list of options for each question, the system chooses the best answer from all potential spans in the passage, which means they must deal with a vast number of possibilities. Spans have the extra benefit of being simple to evaluate.


​​The general construction process is as follows:

  1. Obtain a large number of questions with answers in a specific field ( a standard question set).
  2. Use the BERT model to convert these questions into feature vectors and store them. Then we assign a vector ID to each feature vector at the same time.
  3. Store these representative question IDs and their corresponding answers in PostgreSQL.

The QA setting, depending on the span is extremely natural. Our Open-domain QA system can typically discover the right papers that hold the solution to many user questions sent into search engines. The task is to discover the shortest fragment of text in the passage or document that answers the query, which is the ultimate phase of “answer extraction.”

Developed by Sigmoidal portal has a wide range of possibilities, that include:

  • Full discovery and access to content from PubMed and other high-quality sources
  • We implemented user authentication, that ensures that every search in the library’s collection results in speedy and accurate retrieval of the appropriate sources
  • Unified index – consolidated index of local, remote, and library-specific metadata from a variety of sources presented in a single interface

How did we boost the project with Sigmoidal DNA?

Design Authority

We employed our Data-centric AI approach that involved building an AI system with a strong focus on quality data to ensure that the data clearly conveys what the AI must learn.

Product-focus Method

During this project, as we dealt with massive data sources, we utilized our Early detection of potential pitfalls method, focusing on possible stumbling blocks, the most critical being sampling bias, collecting a perfectly unbiased dataset to counteract any challenges.

Product-focus Method

As we strive for continuous improvement of the product we deliver, Sigmoidal created a model that predicts classifying medical data into different categories. We wanted to ensure that the product would be market-ready and easy to improve upon in the future.

What technical stack has been designed and implemented?