Dear RASA Support Team,
I am part of a Project Management team overseeing the financials for a diverse clientele. Our current initiative involves constructing a chatbot solution for a major client to streamline their financial data management. The client’s financial data is currently available in CSV format, which our end users manipulate to generate reports and respond to stakeholder inquiries. Our goal is to simplify this process by enabling users to interact with a chatbot to obtain the necessary information.
Although we are not highly skilled in programming, we have been utilizing ChatGPT to assist with Python coding tasks. We aim to pilot the RASA open-source platform with a user base of approximately 200-250 individuals to evaluate its potential in reducing their workload. Should this trial prove successful, the client’s IT/Infrastructure team would take over for broader implementation. Additionally, a positive outcome could lead to the adoption of RASA chatbot solutions by other clients we serve.
Enclosed, please find a sample “chatbot_data” file and a "chatbot_questions chatbot_data.csv (17.1 KB) chatbot_questions.csv (5.1 KB) " file. While these files do not contain sensitive company information and are not representative of our actual data, they will serve as a basis for understanding how to construct the chatbot. The data file includes headers on row 2, with columns R to AD representing financial data for FY-24 and columns AE to AQ for FY-25. The term “Actuals” refers to received invoices, while “Forecast” pertains to projections for current and future months.
Our data structure includes hierarchical levels such as “Org Name,” “Enterprise,” “Portfolio,” “Product,” and “Application.” We are seeking guidance on the following:
- Intent Design: We are struggling to conceptualize the intent file, as each question seems to be a potential intent. RASA documentation advises against creating too many intents, so we need advice on how to approach this.
- Database Information: We need to store certain data, such as approved and released funds for various financial years, in a database. We are considering using “slots” to manage this information, but we are open to suggestions for better methods.
- Handling Acronyms: We anticipate scenarios where users might employ unofficial acronyms that we do not use. We seek advice on how to address this in the chatbot’s design.
Currently, we are in the brainstorming phase and have installed RASA. As we progress, we may have additional questions. We also plan to integrate the chatbot with Streamlit to provide visual representations like graphs and charts for user queries. While we understand that this aspect may be beyond the scope of support, any guidance relevant to the chatbot’s design would be appreciated.
We look forward to your expert advice to help us navigate these challenges and create an effective chatbot solution for our client.