Healthcare is one of the pioneer industries to adopt the chatbot technology. Normally, all people deal with healthcare issues on a regular basis. Hence, there is the need for any healthcare service provider to keep its customer service outstanding to retain its customers. Many medical institutions are now served by chatbots as a customer service 24 hours in a day. To achieve this without excessive reliance on manpower, ServReality has developed SwiftAnswer.
ServReality designed the SwiftAnswer to be a customer service assistant, with the ability to serve even the largest medical chain. It is powered by the natural language processing technology. Made for a mini healthcare institution, SwiftAnswer can be used to reduce reception workload, by providing patients with swift answers to their requests at any time given. The chatbot helps patients get fast and correct answers to their questions. From requesting the nearest drug prescription institution to Wi-Fi password, SwiftAnswer sends the required information to the Gmail address of the guests.
SwiftAnswer was created by the ServReality team consisting of a full-stack developer, a QA specialist, a Machine Learning/Natural Language Processing (ML/NLP) specialist, as well as a project manager. The timeframe for its completion was two months. Technologies incorporated in its development include:
- Python (Flask)
- Gmail API
- Api.ai/Watson IBM
- Gensim (Word2Vec)
- Beautiful Soup.
The SwiftAnswer consists of three parts:
- The chatbot itself
- The knowledge store
- The administrative panel. The administrative panel is for the effective management of the bot knowledge, information and documents storage, and monitoring its conversations.
The key functionalities of SwiftAnswer include:
- Receiving and processing requests from users
- Giving relevant and quick response, based on the stored data present in the knowledge store
- Looking for documents in the database, to find the right answers to questions by the user.
The project development process began with writing the technical specification. After this, prototypes were prepared for the admin panel. It took about two weeks to put the necessary things in place, before commencing the actual development process. The project was implemented in two months, with the team described above. Using Scrum-ban – an event-driven approach to development, the ServReality team of experts worked in short repetitions, with one reputation planned ahead. The team also drafted progress reports at the end of each sprint, in a bid to be quicker and more flexible.
After creation, the chatbot was then trained to expand its capacity further. Ordinarily, the chatbot is designed to process about a thousand letters in 24 hours. If the chatbot understands the details of the request, it will reply instantly. And the person will get the reply in his or her email box with the proper name, alongside an attachment if requested. In the case where the chatbot does not recognize certain keywords or combination of words in the patient’s request, a notification will be forwarded to the human operator, saying “I do not understand this request”. If this happens, the operator will send the reply himself or herself. However, all unanswerable requests are sent to the language knowledge base for chatbot training, hence expanding its knowledge base.
By making use of DialogFlow (Api.ai) and IBM Watson, the chatbot was designed to recognize language conversions. The ServReality team also developed a Python algorithm that searches specific words in the email text and looks through the database for the required document.
Conclusively, SwiftAnswer chatbot does not only bring speed and improved quality to customer service, but it also shed the workload of the healthcare institution staff. ServReality at this moment tips the chatbot to even become more powerful with time, and soon no human operators will be needed.