⚠️ This page is being updated (31/08/2022 - 14:55) ⚠️
In this project I designed and developed, on Cosmobots platform, a back office virtual assistant responsible for two distinct tasks: helping users get rapid access to information and automating corporate card requests. This virtual assistant comes as a solution to a workspace with a large number of employees and collaborators that struggled with disinformation and excessive bureaucracy. HR personnel would often get flooded by e-mails with the same questions over and over, as corporate card requests would take between 2-4 weeks to get processed. 

For the first task, it was necessary to design a FAQ chatbot capable of reading and understanding natural human language (NLP - Natural Language Processing). In order to achieve that, I gathered all available information from different sources (manuals, guides and the company's website) and  built a compilation database. The next step was to make this database easily available so that the bot could easily retrieve any information from there. Using a third-part app I turned an Excel online spreadsheet into a REST API database. This solution allowed us to quickly test our prototype before working with the development team on building a solid and secure database. With this setup ready, the bot was able to answer about any corporate procedures and processes, so it was now time to train the bot to identify the questions. After training the NLP model with Intents and Entities, we had a bot that could understand human language, both text and audio messages. A card carousel menu was integrated within the bot flow, so user can interact by clicking on the icons available on the menu as well.

The video below shows the first prototype in action, working as a FAQ chatbot:
"Asking Gessika" shows how the virtual assistant is able to aid users with their questions about corporate policies.
The images below were produced to introduce the virtual assistant to users, as a quick guide showing which topics are covered and how it works:
The second task was to automate the process of requesting a corporate credit card. Usually, the employee was supposed to print and manually fill a form, then deliver it to the HR staff, which would take between 2-4 weeks to process this request. After interviewing the staff and analysing the whole process, it was noticed that only a small amount of the forms submitted were filled correctly. Since the company already holds the required information regarding each employee, we came to the conclusion that this form could be filled automatically then, which would make it error proof. Once again, we turned a spreadsheet into a REST API, so the applicant just had to inform his registration number in the chat and the bot would be able to retrieve its data immediately.

The only thing still missing was the signature routine: the document should be signed by the applicant and, when approved, by two members of the HR staff. The solution adopted was to integrate a service of digital signature that complied with the company's safety standards. Since Adobe Sign was already in use by the company in other tasks, I managed to incorporate it to the workflow and the whole process became available through the virtual assistant. The processing time, that used to be around 2-4 weeks, was then reduced to 2 minutes.

This video displays a step-by-step on how to request a corporate card in few steps through the virtual assistant.