During May 2024, I presented at the Public Technology Institute (PTI) AI and Cyber Summit in Washington, D.C., and had the opportunity to listen to various impressive talks from local authorities across the nation.
Among the standout practices witnessed at the PTI event – and the most exceptional local government technology initiative witnessed within the previous year – was seen in the “Monty” endeavor in Montgomery County, Maryland.
The Montgomery County presentation crew was led by Shayna Taqi. Taqi functions as the chief change executive for Montgomery County at the Technology and Enterprise Business Solutions (TEBS) Department. Bringing with her over two decades of experience in steering strategic alteration management schemes, her sector bears the responsibility for applying methodologies to improve comprehension and acceptance of extensive innovation and technological rollouts. Additionally, Taqi is presently spearheading the county’s inaugural generative AI (GenAI) initiative. This innovation has broadened Montgomery County’s customer service accessibility, permitting locals and guests to promptly retrieve information about county amenities from a virtual chat representative on the county’s 311 portal.
Having obtained her B.A. from American University and earning a Certified Public Manager designation from the Metropolitan Washington Council of Governments within the Regional Executive Development Program, Taqi proceeded with her interview endeavors for this blog post post the PTI session in quest of spreading the Monty project’s visibility nationwide and worldwide.
Dan Lohrmann (DL): Elaborate on the genesis of the Monty project. What was Monty 1.0 inclusive of during its inception?
Shayna Taqi (ST): Our inaugural website chatbot, Monty 1.0, was initiated in January 2021 to assist the county’s 311 crew in managing the escalating resident service demands throughout the COVID-19 outbreak. Operating Zammo.ai artificial intelligence technology and intricate dialogue structures in English and Spanish, the Monty 1.0 chatbot catered to citizens’ information inquiries touching upon roughly two dozen of the most popular 311 subject matters.
DL: Post-pandemic, what transformations occurred within the project vision?
ST: Amid the extensive wait durations for Montgomery County residents attempting to reach an MC311 customer service agent during the COVID-19 crisis, we identified the necessity for an extra communication avenue urgently, given the time-sensitive nature of many of these requests, such as housing and financial aid. Within merely a couple of weeks, the Monty 1.0 chatbot was set up and went live on the MC311 website through the efforts of a modest project group within TEBS, and via collaborations with Microsoft and Zammo.ai. Monty 1.0 encompassed the primary services frequently inquired by residents. In addition, we altered the launch settings on the telephone system for rapid routing of urgent calls. Within a two-month timeframe from launch, hold times plummeted from 5 minutes to 2 minutes, and call abandonment rates decreased from 25 percent to 11 percent.
Despite efficaciously managing some of the amplified service demands posed on the county’s 311 department, user feedback relating to the Monty 1.0 chatbot frequently leaned towards the negative as it could not tackle all queries addressable by a customer service rep. The procedure for enlarging Monty’s knowledge repository was labor-intensive and demanded advanced technological adeptness, constraining 311’s speed in evolving and sustaining a knowledgeable chatbot. Consequently, residents pointed out that the chatbot’s scope was too restricted and incapable of grasping basic phrases and requests. In direct response to the resident feedback, the county’s Technology and Enterprise Business Solutions division kicked off a proof of concept in May 2023 to scrutinize the utilization of generative artificial intelligence technologies as a remedy to these setbacks.
DL: What capabilities does Monty 2.0 presently deliver? How many languages can you accommodate for residents?
ST: The Monty 2.0 chatbot leverages generative artificial intelligence technologies to facilitate natural conversations with residents; necessitating no intricate dialogue planning. The chatbot can promptly respond to information requests spanning over 3,000 311 subject matters – a hundredfold jump from the chatbot’s previous iteration. Conversations are supported across 140 languages, encompassing the seven most commonly spoken languages in Montgomery County (English, Spanish, French, Mandarin Chinese, Korean, Vietnamese, and Amharic).
DL: What GenAI and other technologies back Monty’s development?
ST: The Monty 2.0 chatbot is grounded on a synapse pipeline (a series of activities executing data-driven procedures) streaming 311 knowledge base data to the county’s Microsoft Azure platform. Here, the 311 knowledge base data is converted and indexed for usage by Microsoft’s GenAI Cognitive Search service. Zammo.ai tools oversee the chatbot application, workflows, and user interfaces.
Simply put, an inputted request is initially received and relayed to the Microsoft Cognitive Search service via Zammo.ai workflow. The Microsoft Cognitive Search service arranges and selects indexed 311 knowledge base details, as per the inputted request. This data is then channeled to OpenAI via Zammo.ai workflow, where it is composed into a coherent response utilizing ChatGPT GenAI technology. The response is then funneled to Zammo.ai and dispatched to the resident, finalizing the request cycle. The entire interaction process occurs within seconds.
DL: How can Monty be accessed? Outline the central advantages Montgomery County residents receive through this service?
ST: The Monty 2.0 chatbot is accessible via the county’s 311 website. The chatbot permits Montgomery County residents to swiftly acquire headline information on myriad county services across multiple languages without the inconvenience of sifting through piles of data or connecting with a customer service rep. Locals simply input or dictate their query within the chat interface, and Monty 2.0 handles the rest. The Monty 2.0 chatbot excels at addressing clear and concise requests, permitting the county’s 311 customer service reps to delve into and troubleshoot more intricate resident issues.
DL: What feedback has the user community provided thus far?
ST: Post its rollout in March 2024, Monty 2.0 has garnered generally favorable reception, as per the analytics compiled by the county through Zammo.ai feedback tools. Resident feedback is automatically elicited using thumbs up/down buttons and comment forms within every user chat with the chatbot, and the Monty 2.0 chatbot’s utilization has more than doubled that of Monty 1.0, concerning year-over-year figures. Generally, locals have welcomed the augmented scope and functionality of the chatbot compared to Monty 1.0; user feedback has also been pivotal in identifying previously undetected flaws and other feature suggestions. TEBS remains committed togather insights from feedback given by residents and continuously improve the Monty 2.0 chatbot by making enhancements to prompts, workflows, knowledge base material, and other backend elements, aiming to decrease the occurrences of inaccurate and unanswered inquiries from residents.
DL: Would you kindly share further details on the feedback from residents?
ST: Resident feedback primarily revolves around three main aspects: enhancing our knowledge base, refining the user experience of the chatbot, and fulfilling requests for new content and functionalities. This feedback has been instrumental in assisting our team in identifying errors and flaws within the knowledge base, such as broken links, outdated information, and related issues. By leveraging resident feedback, we are directing our efforts to enhance prompts on the backend and enhance the performance of the chatbot’s extensive language model infrastructure. We are also maintaining a queue of enhancement requests from residents to guide future updates of the Monty chatbot.
Among the most frequent enhancement requests are features requested to be integrated directly within the chat window itself, such as submitting service forms to county agencies through the chatbot, transferring chats to a 311 customer service agent, and receiving location-specific service details.
Potential exploration into including other service tiers like 911, state, and federal assistance is on the horizon as Montgomery County continues striving to advance service delivery for its residents. Initial steps are also being taken to revamp the county’s web pages.
DL: During your presentation at the recent PTI AI and Cyber Summit in Washington, D.C., your approach of involving your team in presenting various project aspects was commendable. Their enthusiasm for their roles and pride in their work was evident. How did you foster such a positive team culture? And do you have a mix of staff from public and private sectors?
The professionals comprising the Monty 2.0 Project Team include:
Gail Roper, Chief Information Officer and Program Sponsor
Shayna Taqi, Leading the Project
Manu Daniel, Technical Project Manager
Michael Zanfardino, Functional Project Manager
Tushar Parekh, Developer
Sri Harsha Kotagiri, Developer
Colin Cox, Product Owner
Skyler Grubbs, Overseeing Change Management
Lauren Der, Managing Communication and Marketing
ST: The development of the Monty 2.0 chatbot was an in-house effort led by TEBS, with close collaboration from our vendors at Microsoft and Zammo.ai. The project team built a strong bond over the span of 10 months, from the initial proof of concept to the chatbot deployment. It was one of many parallel projects for team members who were temporarily sourced from various divisions across TEBS to contribute their expertise in project management, business analysis, and software development. Hence, we opted for an agile, sprint-based methodology to maximize team efficiency.
Our sprints were intensive, involving extensive collaboration among team members to design, develop, and rigorously test chatbot components. This testing phase comprised numerous testing iterations and a resident focus group to shape our end product. Throughout the process, team members took on different roles as needed, embracing a culture of continuous learning from mistakes and a proactive attitude towards innovation. This diversity in thoughts and skills, coupled with a receptiveness to constructive feedback, was pivotal in the project’s success and is reflected in the strong team camaraderie.
DL: What lies ahead for Monty? What are your projections for the upcoming one to two years?
ST: Our team is currently focused on significant enhancements for the Monty 2.0 chatbot, which include integrating county GIS features to provide personalized information to residents, such as voting center locations, updates on snow clearance activities, and related matters. Collaboration with our Microsoft and Zammo.ai vendors is ongoing to enhance the chatbot’s performance through backend upgrades. These improvements encompass the implementation of Microsoft Azure Prompt Flow, a tool to evaluate and enhance the accuracy of prompts used in guiding the Monty 2.0 chatbot’s GenAI components. Additionally, our product ownership team continues to assess UI/UX enhancements and knowledge base updates to address resident-reported issues comprehensively and boost chatbot adoption. Finally, we aspire to use the expertise gained by our team in other GenAI projects within the county setting.
Since the project inception, we have recognized the significance of continuous improvement and leveraging the operational efficiencies gained during the COVID-19 pandemic. We take immense pride in our team’s capability to swiftly support residents while modernizing our 311 technology and enhancing Montgomery County’s service delivery model and capabilities in a more holistic manner.
About Author
