Free Robotic Process Automation (RPA) using TagUI-AI Singapore project for Student Learning Space (SLS)
Project Summary
TagUI - AI Singapore’s robotic process automation (RPA) project, has reduced the Student Learning Space (SLS) team’s manual effort on repetitive tasks, like crawling data from SLS lessons, assigning HQ officers to review lessons based on their curriculum subjects, emailing officers with personalised email body and sending reminders after lessons are detected as unreviewed for more than 5 working days.
The RPA is written in human-readable code and when combined with JavaScript and Python, it becomes a fully automated flow of commands. TagUI logs in to SLS and saves lesson data on computer, the JavaScript assigns reviewers to lessons, and Python copies selected data back to a Database-Google-Sheet etc, with one schedule-able click .
We have open-sourced our “code-in-words” on GitHub and it is easier for MOE to integrate the use of RPA into other rules-based workflows using WOG (recently whitelisted and added codes), SSOE or personal computers (Windows, MacOS and Linux).
Innovativeness of solution
a)
We were initially given the task to assign SLS Community Gallery (CG) lessons which were pending review to 150 colleagues via emails. We used to do it manually but after reflecting on the experiences and feedback from 2020 when large numbers of SLS CG lessons needed to be reviewed urgently due to COVID-19 disruptions and Home-Based Learning (HBL), we realised there was an opportunity to develop a more scalable approach to save time and effort.
There were some earlier efforts to use UiPath, a leading commercial RPA tool. Having no prior experience using that tool, we still tried to make UiPath run the automation but to no avail. There were no proper documentations to refer to, debugging was difficult and the officer who created the flows had left HQ. After trying for two months, we decided to abandon this UiPath approach which also cannot worked after Jan 2020 after SLS had a major user interface change.
b)
Our TagUI solution is innovative and bold because of the following reasons. Our TagUI script is written in natural language-like syntax in English and can be easily understandable and converted to 19 other languages (Chinese, Malay etc) which supports global usage. The same script-flow can also be written in any code editor like Visual Code Studio or even Microsoft Word and Excel via the TagUI plugin for novice users. One can simply run the flow via terminal or command line or double click a ‘deployed’ file. It also runs on all three operating system MacOS, Linux and Windows, supporting MOE to scale up the use easily unrestricted by computer operating system. It is also easy to install, simply unzip, add path and run, assuming the dependencies are already on the computers. We also tap on JavaScript and Python integration for tapping on other libraries for more complex features that TagUI doesn’t do well. Finally, TagUI is free and open-sourced, so MOE can continue to use it perpetually unlike other RPA tools that may cost approximately \(420/month (e.g. UiPath) to maintain per account.
We started by writing simple lines of code in TagUI and gradually added more complex JavaScript to supplement beyond what RPA does, like assignment of reviewers to lessons automatically based on reviewers’ 2 subjects-interests and levels (Primary or Secondary). Python was used to do file manipulation and copying of data back to the lesson data Google Sheets intelligently at the desired columns and rows. Having experienced success in RPA, we extended the flow to include searching emails and replying to gently remind reviewers to login to SLS to review the CG lessons, after detecting that the reviewer was taking longer than the Public service commitment of 5 working days. We were able to do so much because of the technical support from the TagUI AI-Singapore team whenever a “blocker” is encountered and we were able to overcome the issues very quickly via GitHub issues -discussions, telegram community chat , or Zoom sessions with the TagUI maintainer(s)-creator(s).
Idea generation and development
a)
After failed-experimenting with UiPath for two months, we decided to take the opportunity to find a free and open-source tool which could hopefully do what we wanted to address the abovementioned challenge. Since TagUI is listed as an AI Singapore RPA project, we decided to check it out by ‘installing’ it into our personal computer. We tried to run some of the sample ‘.tag’ files to understand how it works and we were finally able to write simple RPA flows. We began to ask questions via the GitHub issues and tried to make simple flows work using text files with a ‘.tag’ extension. We took calculated risks and showed spirit of dare by using an AI Singapore, open source RPA tool, knowing that there will be resistance from strict MOE IT policy on running TagUI on Whole of Government (WOG) devices and School Standard Operating Environments (SSOE ) computers thus we did it on our personal computers. We had some prior experiences succeeding in using Open-Source Physics tools so the digital transformation benefits outweighed the risks of being seen as ‘unauthorized’ use of RPA tools. The time spent trying to learn UiPath and TagUI also helped as we were clearer about the lingos, terminologies and missteps so we applied the learning from the failed UiPath attempt.
b)
We knew from feedback that the reviewers wanted customised emails with hyperlinks to exact CG lessons to ‘return/feature’ thus our RPA captures the URL of the CG lessons and adds it to the emails. Reviewers also requested gentle email reminders just in case they forgot about the review by the due date.
We collaborated with the TagUI AI Singapore team (Ken and Ruth) to help with some of the syntax, technical know-how and even trouble-shoot how to run TagUI on SSOE Windows computers. We are in the process of even whitelisting TagUI and Python for use in WOG and SSOE, because we believe in the tremendous potential to automate rule-based clicks to save human efforts in clicking on webpages and desktop apps for MOE teachers and officers.
Separately, this project has also automated adding of test questions (Multiple Choice and Free Response) drawing the data from a CSV file , to support population of question bank for Primary Mathematics for CPDD’s adaptive learning system (ALS) efforts. Another automation is to add 170 teachers from an Excel Sheet for ETD’s professional development workshops, with ability to skip if the teacher’s email is not found, , estimated clicks, 170 teachers x 10 clicks = 1700.
For scalability across MOE, we created YouTube explainers and working scripts-texts (without sensitive data ) to help other MOE colleagues to learn so that anyone in MOE can “take over” the project and run the workflow or adapt them to suit different MOE work-streams. We even volunteered to help with MOE Olive Fiesta 2021 fund-raising efforts by showcasing our RPA flows as “magic” to excite and inspire.
Benefits and impact to stakeholders
Benefits to MOE, Cost Savings
Building such a full automation on SLS front-end itself might have costed an estimated \)80,000 upfront and $5,000 for yearly maintenance. Besides, SLS funds are strictly for meeting high number of user needs (35,000 teachers and 500,000 students) for higher return of investments. Spending funds on a time-saving feature like our project would have impacted only 150 MOE-HQ officers serving as SLS CG lesson reviewers, and thus, unlikely to be prioritised for a built soon. This gives a strong case for RPA to come into this part of the SLS workflow.
Benefits to CG Chief Admin:
a) Automate Mundane and Rule-based Work
Direct benefits are for the members of the project who have to do the following routine tasks: get the CG lessons data, assignment of CG lessons to reviewers if it is not assigned yet, send emails to reviewers with instructions and URL for quick reviews, send email reminders for reviews than are still in the pending review tab after 5 days after the teacher’s submissions.
b) Life-Long Digital Learning
Instead, we used our digital literacy skills in coding, and automate all flows into a single click on the computer. Of course this requires more digital skills, but we argue it is far better a learning experience for us to use our brain to problem-solve and code than to do mundane manual clicking and rule-based assignments.
c) Higher job satisfaction and assignment-email with accuracy and fairness
Estimated 1500 CG lessons were assigned out automatically using TagUI from Apr to Dec 2021, on every work-day basis (190 times), and 8 emails approximately every time, with auto email reminders (2 emails daily). If the team had to do all these manually by hand, the job satisfaction would be very low. In addition, the accuracy of the emails and fairness of assignment of lessons to reviewers has been given far better results than if done by hand. We estimate a time saving of 30-60 minutes per day depending on the number of new pending review lessons that come in to CG of SLS.
Benefits to 150 CG lesson reviewers
Our CG lesson reviewers have received personalised emails to ask them to click on a single SLS URL link and can start their review process. The RPA also send emails after detecting number of working days greater than 5 days after teacher’s lesson submission date, to gently email reminders them to remember to do the lesson review by the next working day.
Sustainability of solution
Zero dollars and always Evolving/Expanding using GitHub
The solution would be sustained beyond its implementation phase using zero dollars through an ETD Lead Specialist’s technical know-how. We are in the process of exploring a WOG TagUI whitelisting (using This email address is being protected from spambots. You need JavaScript enabled to view it. instead of a generic email address This email address is being protected from spambots. You need JavaScript enabled to view it. ) Update!! Possible after whitelisting by ITD . With each automation made, we will publish the file *.tag and other dependencies files *.csv, which is just text and code and YouTube URL, to share the source codes with MOE officers on GitHub.
Sharing at MOE-Charity Event
We also shared about TagUI capabilities during the 2021 Sept 09th OLive Fiesta SPICE Carnival - Talent Showcase MOE - Robotic Magic TagUI. The participants were very excited about TagUI’s computer and web automation and we believe we helped MOE colleagues to be aware of the free RPA can offer in terms of higher accuracy, savings in time and cost, ease of writing code etc.
YouTube TagUI MOE playlist for scaling
We created YouTube playlists of videos and working scripts-texts (without sensitive data ) to help other colleagues in MOE to learn and share knowledge. We are also continuously experimenting and exploring other use cases of TagUI for routine MOE work streams and we will prepare video tutorials so that anyone in MOE can adapt the project and run the workflow to suit different work requirements.
Scalability to other MOE-SLS work
We also automated the adding of test questions (Multiple Choice and Free Response) drawing the data from a CSV file, to support population of question bank for Primary Mathematics for CPDD’s adaptive learning system (ALS) efforts.
Additionally, we automated adding 170 teachers from an Excel Sheet for ETD’s professional development workshops, with ability to skip if the teacher’s email is not found, has saved a lot of mundane – 170 teachers x 10 clicks = 1700 clicks.
Lastly, we also automated the sharing of lessons in SLS “My Drive” to different teachers for co-creation of lessons or activity templates, which becomes painful when sharing 20 lessons to 20 teachers = 400 sharing x 10 clicks = 4000 clicks.
Almost limitless RPA for Desktop Apps and Chrome Browser
The longer-term plans for the project could be to automate all other mundane rules-based clicking on SLS, and the project can also be the MOE community champion for use of TagUI-RPA for MOE’s digital transformation efforts, such as clicking using computer vision clicks on computer desktops apps and Chrome browser webpage automation etc.
TagUI RPA Professional Development Community-Champion
The team is also ready to support others’ Professional Development in the use of TagUI as well as point them towards readily available resources including reference documentation, a free beginners and soon-to-be-ready intermediate online certification course to pick up the use of the open-source tool without the need for prior programming knowledge. We strive to help all MOE colleagues to digitally transform their own repetitive workflows where possible, using this free RPA and “Learn for Life” for higher job satisfaction. We can also share on Workplace to help build a community of RPA enthusiasts and transform work processes together.
Please attach all your Annexes and supporting documents (e.g. photos, links to website, graphs, etc) here, where applicable.
Photo of Sharing Event of 20210909 OLive Fiesta SPICE Carnival - Talent Showcase MOE - Robotic Magic TagUI a free tool for Robotics Process Automation (RPA) to 50 participants from MOE.
Featured on
https://github.com/kelaberetiv/TagUI as a demonstration of sustainability of our solution goes beyond MOE to scale. Lawrence Wee – Second Top Left.
WhatsApp discussion with the ex-ETD officer who use UiPath extensively, who also won an innergy award in 2020 for the work on UiPath-RPA, suggesting why TagUI is “more stable” (mostly a text *.tag file and *.csv) while UiPath is “slow and buggy with lots of overheads due to graphics” for computer vision clicking.
Screenshot of RPA email to reviewers with personalised title and clickable URL in table format, easy for lesson reviewers to do their reviews
Screenshot of RPA email gentle reminder that is trigger after detection of greater than 5 working days after teacher’s lesson submission date from day of email sent. The RPA searches the email and reply all, ensure all data is intact.
Screenshot of email from one of the reviewers that they missed out the first assignment email and appreciate our “streaming process” such as RPA lesson review, publishing CG e-Catalogue of all 7000+ CG lessons that we can also RPA out from SLS webpages etc.
