Tech Demo by Nakisa (LPL1-V76)

Transcript: Tech Demo by Nakisa

[00:00:06 Text appears on screen: Tech Demo by Nakisa.]

[00:00:12 The screen fades to Wendy Bullion-Winters.]

Wendy Bullion-Winters (Vice-President, Business Enablement & Chief Financial Officer): Welcome. I'm happy to welcome you today to another event that is a part of our Canada School of Public Service Tech Demo series. My name is Wendy Bullion-Winters. I am Vice-President of Business Enablement and Chief Finance Officer here at the School. And today, I'm happy to be joined by Nitin Sharma, Director of Product from Nakisa.

[00:00:40 Nitin Sharma appears in a separate video chat panel.]

Wendy Bullion-Winters: This is a company born in Montreal that maybe some of you might recall. They've been operating for over 20 years and you may be familiar with some of their products, especially their organizational design tool that worked synchronicity with PeopleSoft to create live and up-to-date org charts. Today, Nitin is here to tell us about another really exciting innovation by this Canadian company and how this company has been experimenting and building more complex tools that leverage AI. So, Nitin, welcome, and over to you.

Nitin Sharma (Director of Product, Human Resources, Nakisa): Pleasure. Thank you, Wendy, and thank you, everyone, for your time. And as you mentioned, Nakisa is a Montreal-based company headquartered and founded here. We focus on, of course, enterprise software throughout that time. We've grown to be roughly 300 employees. We have proven deployments and large enterprise customers globally. More specifically, we're very proud of the work we've been doing in the public sector, namely with the Federal Government of Canada, Ontario Power Group, and multiple other provincial authorities. And at the same time, Nakisa does also have its Canadian government Protected B security clearance. So, that means not only personnel clearance but everything that goes around the adherence of federal government requirements of hosting, data residency, compliance, security, so very much looking forward to today's conversation and discussion.

Wendy Bullion-Winters: Great. So, today, you'll be giving us a demonstration of Nakisa Decision Intelligence that I believe you launched in 2025. This tool, I had a quick demo of it, but I'm excited to see more today. I understand this tool can be used to interpret data of any kind, for example, workforce data, financial data, or even just client databases, and you're going to show us more about the power of this tool and how it unifies and interprets and analyzes a lot of internal and external data in real time, how it runs forecasts, it can do simulations, and provides analytical insights. Even more, you can get the tool to use natural language and asking complex questions to get tailored advice and recommendations based on the data. So, Nitin, for today, I understand you're going to use, to demonstrate the tool, you're going to use some open source data from the company, the Montreal company, Bixi. So, for those in the audience who are not familiar with Bixi, it's a company in Montreal that's run. It's a public bike-sharing system. So, essentially, they have various docking stations across the city where you can rent a bike using a credit card and linked through their Bixi app. It's a really popular service and they have ridership data that they make publicly available. It's open source. So, it's a really excellent dataset to be using to demonstrate the power of this tool, this Nakisa Decision Intelligence tool. So, Nitin, over to you.

Nitin Sharma: Thank you, Wendy, and that's right. I thought it would be a good idea to run through a publicly-available data source to show off Nakisa Decision Intelligence's capabilities, and the discussion of, why are we looking into Nakisa Decision Intelligence, is two parts. There's been the Government of Canada's directive on automated decision-making, so looking for artificial intelligence to support administrative decisions and to overall improve delivery service, and that aligns with many of our conversations where organizations are moving beyond static reports or interactive dashboards, that they're really looking for ways to make transparent and explainable AI decisions where humans are kept in the loop and are in the driver's seat, but ultimately, that they are able to rapidly interact with the data, gain insights, run what-if scenarios and simulations, and the Bixi dataset we thought was a great way for us to show off these capabilities, ultimately because it's a dataset that has seasonality, it has demand spikes, it has a workforce constraints. It's not just about moving bikes from one location to another, but there is everything around service readiness, staffing, accountability, and that's the reason we'll run through the Bixi dataset today as we look at Nakisa Decision Intelligence here.

Wendy Bullion-Winters: Perfect. Yeah, I think it's something that our audience will be able to really envision. Everyone has, at one point or another, wanted to… or maybe not, but I mean, we can envision wanting to rent a bike and take it from one location to another. So, I think it's a really great use case today. So, you describe Nakisa Decision Intelligence as agentic. This is a word, a little buzzword, that we hear now and then when we're talking about AI, this word "agentic". What does this mean, really?

Nitin Sharma: That's a great question, and agentic could mean different things to different organizations. And the way we look at agentic capabilities within Nakisa Decision Intelligence, so you see our application over here, and where I actually want to begin is by looking at the data.

[00:06:34 A webpage is shown, displaying dataset information for Bixi bike usage in 2024 and a description that reads: "Contains detailed records of Bixi bike trips, including start and end station names, locations, arrondissements, and trip start/end times. Useful for analyzing bike usage patterns and station activity – 10 fields." A list of sample questions are shown under the headings: "Demand & Usage Trends", "Time-of-day & Seasonality", "Station Performance & Network Optimization", "Spatial Flows & Coverage", "Trip Duration, Distance, & Speed", and "Operational Planning & Rebalancing".]

Nitin Sharma: So, data is made available to NDI. That can be connected, as you had mentioned before, directly to an ERP, an Oracle, an SAP, whichever major ERP there is out there. This data can be uploaded through CSVs, or most common challenges public sector or enterprise faces, the data lives in multiple systems. So, you can connect with multiple systems as well. And once you have the data, what agentic means for Nakisa Decision Intelligence is, number one, you don't need to pre-train anything.

[00:07:06 Nitin Sharma clicks "10 fields" and a list of fields is generated.]

Nitin Sharma: And with the data available, Nakisa Decision Intelligence is able to look at the fields and add the context around them. So, it didn't need to be trained on anything to be able to do this. And if you have data coming from multiple systems, it can also connect all of that data together automatically. So, you have an agent that understands data and context, another agent that connects data across these different systems, and another agent that actually says, with this data and how it's connected, here are the types of insights that we can surface right away, which you might not have been aware of. So, when we talk about Nakisa Decision Intelligence, for us agentic means this orchestra of agents that gets your data, understands your data, connects your data, and then surfaces insights on the data without needing to be trained on anything in advance, and that all of this happens seamlessly in minutes.

Wendy Bullion-Winters: That's amazing. So, before we dive into the Bixi public data and kind of get the demonstration today, I want to kind of take what you said and interpret it for a government example, and I'll take my own example. So, I run all of the corporate services here at the Canada School of Public Service. So, I can imagine using this tool, putting in some PeopleSoft data so that it understands my workforce, putting in some SAP data so it understands my financials, and then also putting in maybe some learner data, the types of courses we deliver, the attendance rates, the feedback from the faculty, the dropout rates, the no-show rates, and you're telling me that Nakisa, that your tool here, would be able to interpret, using this agentic intelligence, would be able to automatically determine the context, understand that this is a learning institute, understand that we're talking about course delivery and learning delivery just automatically. I don't have to do anything more than just upload the data itself.

Nitin Sharma: That's right. It would understand the data, show you what it understands around the context of the data, show you how the data connects together. And then, with that, with no pre-training, is able to let you know how can you interact with it, what types of insights can you surface from it? And even if you have questions based on data that you don't have, it won't hallucinate and make up responses. It would actually let you know, to answer this question, what type of data would I need? So, that can also enable you to go connect either with other systems or go augment the data that you had provided to get key insights.

Wendy Bullion-Winters: That's such a great point to have raised because I think we do have concerns, if we've ever dabbled with certain AI interfaces about this phenomenon of hallucinations. So, thanks for that. Okay, well, without further ado, let's dive into some of these insights. I'm just looking.

[00:10:27 A webpage is shown, displaying dataset information for Bixi bike usage in 2024 and a description that reads: "Contains detailed records of Bixi bike trips, including start and end station names, locations, arrondissements, and trip start/end times. Useful for analyzing bike usage patterns and station activity – 10 fields." A list of sample questions are shown under the headings: "Demand & Usage Trends", "Time-of-day & Seasonality", "Station Performance & Network Optimization", "Spatial Flows & Coverage", "Trip Duration, Distance, & Speed", and "Operational Planning & Rebalancing".]

Wendy Bullion-Winters: And okay, so time and day and seasonality, station performance, spatial flows and coverage. Can you click on what month recorded the highest total number of trips?

Nitin Sharma: For sure. Actually, we'll jump into NDI.

[00:10:44 A new webpage appears with a chat function.]

Nitin Sharma: So, they're always showing the breakdown of the data, how everything was interpreted, and let's jump in. So, we're in NDI right now.

[00:10:55 Nitin Sharma types "What month recorded the highest total number of trips?" into the chat box and clicks the send button.]

Nitin Sharma: And you had asked what month recorded the highest.

Wendy Bullion-Winters: The highest total number of trips.

Nitin Sharma: Total number of trips.

Wendy Bullion-Winters: I want to know, when is Bixi the busiest? When do they rent the most amount of bicycles?

[00:11:04 Text appears on screen:
"Uncovering insights from your data, please hold tight."
"Transforming raw data into actionable intelligence, just a moment."
"Visualizing patterns and trends for your analysis, hang on."
"Exploring data relationships to reveal hidden gems, almost there."
"Crunching numbers to deliver clear insights, thank you for your patience." "Generating comprehensive visualizations for your data, please wait."]

Nitin Sharma: And you can see, NDI is analyzing the data, interpreting the question, looking at the data and the context to come back and let us know.

[00:11:16 A response is generated which reads: "July 2024 recorded the highest total number of Bixi bike trips, with over 2 million trips taken that month. This likely reflects peak summer usage when weather and events drive mobility." Sample follow-up prompts are also provided:
"Show me the month-by-month trend for 2024"
"Which day of the week is most popular for trips?"
"Can we see usage patterns by arrondissement?".]

Nitin Sharma: So, July was the highest number with trips and it had over two million trips that month that were done.

Wendy Bullion-Winters: Okay, and so this public dataset is just for the year 2024. Is that correct?

Nitin Sharma: That's correct.

Wendy Bullion-Winters: Okay. So, we're just looking at 2024 and they're telling us July. Well, that doesn't surprise me. It is the most beautiful time to visit Montreal, and you can imagine a lot of tourists wanting to go around on bicycles in the city. So, now, let's think about, could you ask it to show the most popular docking station?

Nitin Sharma: Absolutely.

Wendy Bullion-Winters: It's actually suggesting things here, "Show me the month-by-month trend." So, that's really cool. "What day of the week is the most popular for trips?" and "Can we use usage patterns?" So, it's already proposing additional questions that I could drill into, following the question that I asked. So, instead, why don't we click on the month-to-month trend before we ask my next?

[00:12:15 Nitin Sharma clicks on "Show me the month-by-month trend for 2024".]

Nitin Sharma: And you're absolutely right there, Wendy, as we're interacting with Nakisa Decision Intelligence, we're in one chat, and here, we're focusing on Bixi data, but you could have different chats because you might be focused with that dataset on different items.

[00:12:28 A response is generated which reads: "Here is the month-by-month trend for Bixi bike trips in 2024, displayed as a line chart. This view highlights how usage fluctuates throughout the year, making it easy to spot peak and low months. Would you like to dive deeper into factors driving the monthly trends, compare specific months, or see the data behind the chart?" A line chart showing Bixi bike trips per month in 2024 is shown on the right-hand side of the screen. Sample follow-up prompts are also provided:
"Show the underlying trip data for each month in 2024"
"Compare trip volume between peak and low months"
"Break down monthly trends by day of week"
"Analyze weather impact on monthly trip trends"
"Summarize total trips by quarter for 2024".]

Nitin Sharma: And with the questions that we're asking, it's understanding our context and it's building its chat memory around it. So, because it understood we want to understand number of trips, as you highlighted, the recommendations of follow-up questions will continue down a similar theme. So, now, we ask for the month-by-month breakdown for those different trips versus the kind of trip average that we see as well.

Wendy Bullion-Winters: That's amazing, and it's not surprising that the winter is the least popular time for people to be riding around Montreal in the snow on these rental bicycles. So, if I was trying to make some decisions about optimizing my business for Bixi, I might want to know where is the most popular docking stations. So, can we ask it that? Where's the most popular docking station and how many bicycles were taken from this docking station in this 2024 period?

[00:13:28 Nitin Sharma types "What are the most popular stations in 2024?" into the chat box and clicks the send button.]

Nitin Sharma: Absolutely. So, let's start with what are the most popular ones for 2024, stations now instead of overall number of ridership.

[00:13:43 A response is generated which reads:
"The most popular Bixi bike stations in 2024 by trip count are:

1. Métro Mont-Royal (Utilités publiques / Rivard)
2. du Mont-Royal / Clark
3. Laurier / St-Denis
4. des Pins / St-Laurent
5. Marquette / du Mont-Royal
6. Métro Atwater (Atwater / Ste-Catherine)
7. Métro Peel (de Maisonneuve / Stanley)
8. Laurier / de Brébeuf
9. Émile-Duployé / Sherbrooke
10. Métro Papineau (Dorlon / De Maisonneuve)"

"Would you like to see a map of their locations, analyze their usage over time, or explore what makes these stations so popular?".
Sample follow-up prompts are also provided:
"Show a map of the top 10 Bixi stations for 2024"
"Analyze trip trends over time for these top Bixi stations"
"Break down top stations by time of day usage in 2024"
"Compare total trip counts for these stations in 2024"
"Summarize top factors contributing to high trip counts at these stations".]

Nitin Sharma: And what we can even do then is now visualize kind of the patterns in it. So, this is great. We understand the top 10 busiest stations.

Wendy Bullion-Winters: To Métro Mont-Royal, and Laurier / St-Denis, yeah, Marquette.

Nitin Sharma: I would call out here, so number one, it's great getting these responses, but what's very unique about NDI, Nakisa Decision Intelligence, we refer to it as NDI in case you see me or hear me say NDI, is that with all of the responses that you're seeing, all the visualizations, the intent is never to give you magical answers. You're depending on the data, as you had mentioned, if you have employee data, performance data, salary data, you're looking at sessions. Whatever it is, most importantly, you need to know what is happening underneath the hood for me to get these answers.

[00:14:46 Nitin Sharma clicks on a question mark icon in the chat box and a new webpage appears with a summary that reads: "This calculation identifies which bike stations are the most popular places for people to start their trips. It tells us the 10 stations where the most rides began, showing us which locations have the highest demand for bike rentals."]

Nitin Sharma: So, for every response that you have, you have the ability to view an explanation. You can actually look at the rationale of NDI, how it interpreted the question, the summary of what it did to then give you a response.

[00:14:59 Nitin Sharma clicks on the "Explanation" heading at the top of the screen, and a technical explanation is generated which reads: "This query selects the name of the start station and calculates the count of bike trips that begin at each station. It groups the records by the start station name, so each row in the result represents a specific station. It then counts the number of trips for each station, orders the results in descending order by trip count to get the most frequently used stations at the top, and limits the output to the top 10 stations with the highest number of departures."]

So, now, you have a technical breakdown of how it interpreted the question.

[00:15:07 Nitin Sharma clicks on the "Detail" heading at the top of the screen, and a mathematical detail is generated which reads: "For each value of Start Station Name in the set of bike trip records, we compute the cardinality (size) of the subset of trips that start at that station. We then rank all stations in descending order according to this count, and select the 10 stations with the largest counts. Formally: For each station s, compute N_s = |{trips: Start Station Name = s}|, then select the 10 stations with largest N_s."]

Nitin Sharma: And if it needed to do calculations, the mathematical formula and calculation, and this is available for every single response, so whether that's for key callouts like transparency, accountability, governance, or we need to go back and review, why did we make these decisions? With every single response that we see with NDI, I just wanted to call out that you have the ability to have built in explainability at every step along the way.

Wendy Bullion-Winters: That's amazing, because sometimes we ask a question and we get an answer and we're surprised by the answer, the response that we get, and it could be in the way we formulated the question. So, that's a really useful tool as well as for transparency and accountability. Excellent. So, can you visualize now these top 10 most popular stations?

[00:15:59 Nitin Sharma types "Use a chord diagram to visualize the flow between top 8 districts" into the chat box and clicks the send button.]

Nitin Sharma: We can. So, what I actually asked it here was to give me a specific type of visualization now, a core diagram. Earlier, when we talked about agentic agents, I touched on data and the interpretation of data.

[00:16:16 A response is generated which reads:
"Here's a chord diagram visualizing the flow of Bixi bike trips between the top 8 Montreal districts in 2024. Each curve shows the volume of trips linking one district to another – thicker curves represent greater flows. This helps identify which district pairs have the strongest bike mobility connections. Would you like to explore which districts serve more as origins versus destinations, see trends by season, or focus on a specific pair for deeper analysis?" A chord diagram showing Bixi bike trip flows between top districts in 2024 appears on the right.
Sample follow-up prompts are also provided:
"Show districts with the highest bike trip origins"
"Show districts with the highest bike trip destinations"
"Analyze seasonal trends in bike trip flows between districts"
"Filter flows to highlight a specific district pair"
"Summarize the strongest district-to-district connections".]

Nitin Sharma: What's happening here now as well is another orchestration of agentic agents. So, the question I asked it, an agent is interpreting that and understanding, do I have the data to answer this? Then, it's retrieving the data, and then it's saying, what is the right way to respond to your question? Is that a text response? Is that a response in a table format or a visual? You could, of course, tell it exactly what you want, but that is also an orchestration of agents that's happening.

[00:16:51 Nitin Sharma enlarges the chord diagram.]

Nitin Sharma: So, that's a second way how Nakisa Decision Intelligence is an orchestration of agentic agents. And here, I had said, those top eight stations, visualize, in a core diagram, the start trips to the end trips, the start to the stops.

[00:17:08 Nitin Sharma hovers over a line on the diagram, signifying the route of Ville-Marie to Le Plateau-Mont Royal with a number value of 596,045.]

Nitin Sharma: So, we could actually see the flow of bicycles, how from Ville-Marie to the Plateau is one of the most common.

[00:17:14 Nitin Sharma hovers over a line on the diagram, signifying the route of Rosemont – La Petite Patrie to Le Plateau-Mont Royal with a number value of 452,661.]

Nitin Sharma: And from Rosemont–Petite Patrie to the Plateau. So, the Plateau is very common. It's a hotspot it seems.

Wendy Bullion-Winters: This is an amazing visual because if I'm thinking to make optimization decisions, to decide, okay, where is my cash cows, where is it most popular that I need to put the most bikes, and where are my dogs, where are the places that I might want to cut stations because they're just not making money or the bikes aren't being used? This would be a question that could inform that decision.

Nitin Sharma: If we wanted to visualize it by district, we can follow up with that, and what I wanted to call out is we see that Ville-Marie to Plateau was the busiest.

[00:17:56 Nitin Sharma clicks on the Ville-Marie to Le Plateau-Mont Royal line and a list of Bixi trips on that exact route is generated below the chord diagram.]

Nitin Sharma: But I could actually click directly there and drill down to all of the trips and the data that was flowing from one to the other. So, it is not just giving me visualizations, it's enabling me to interact with them and understand the data behind them as well, which I just wanted to call out here.

Wendy Bullion-Winters: That's amazing. Okay. So, then, we could very quickly, using that visual, identify the outlier, right? We can see the thicker lines are the most popular and the thinner lines are the less popular. So, just on the same thinking, if I wanted to optimize and I wanted, for example, to suggest two locations where I should add new Bixi stations based on current ridership, is that something that this could do?

Nitin Sharma: Absolutely.

[00:18:53 Nitin Sharma types "Where should we add 2 new Bixi locations to optimize our fleet based on current data?" into the chat box and clicks the send button.]

Nitin Sharma: So, let's jump in, and maybe we wanted to know, where should we add these two new Bixi locations to optimize our fleet? And this one's a great question because it's asking to make decisions based on data we don't have.

[00:19:09 A response is generated which reads:
"To recommend two optimal locations for new Bixi stations, let's rely on strategic data-driven criteria:

Approach

• Identify trip "gaps": Look for areas with high trip density but a lack of nearby stations, or frequent start/end points far from current stations.
• Find overloaded stations: Highlight stations consistently handling the most traffic or experiencing rebalancing issues.
• Spot underserved districts: Pinpoint arrondissements with relatively lower station coverage but notable trip initiation or termination from adjacent districts.

Next step: Data-Driven Recommendation – I'll analyze:

• Stations with highest overflows (departure or arrival demand) relative to their neighbours.
• Trip clusters starting or ending far from the 2024 station network.

Let's begin by identifying areas with high trip volume but fewer station options nearby. Once we have that, I'll recommend two precise locations for new stations – supporting a more balanced and optimized fleet distribution. Would you prefer a focus on central districts (like Le Plateau-Mont-Royal or Ville-Marie), outer boroughs, or citywide optimization? If you have specific priorities (commuter demand, connectivity, areas with growth, etc.), let me know so the recommendations fit your strategy."]

Nitin Sharma: We're saying, go understand Montreal, go understand bike ridership without ever being pre-trained on any of that and offer recommendations. So, right away, it's saying, for these two new locations, we could have looked at, well, where do we have the highest overflow or trip clusters that are starting or ending? And this is great. It's giving me the rationale for why it's going to make these decisions, and maybe I want to know very specifically now. Let's get specific. Let's get three coordinates of the locations themselves.

[00:19:56 Nitin Sharma types "For each district, suggest three specific coordinates where new Bixi stations would have the greatest impact on reducing the load on neighbouring stations?" into the chat box and clicks the send button.]

Nitin Sharma: So, for the different districts, give us specific coordinates of where we want to add these Bixi locations.

[00:20:03 An lengthy response is generated.]

Nitin Sharma: And we can see over here.

[00:20:08 Nitin Sharma clicks on one of the sample follow-up prompts, "Show top three recommended coordinates per district with rationale."]

Nitin Sharma: Yes, so show me your top three recommended with the rationale for why you're giving us this answer.

Wendy Bullion-Winters: Okay, that's really neat, and what about if I wanted to… so, right now, it's generating that based on the data available from the dataset, but what about if I wanted to bring in information from the internet? For example, include in your assessment upcoming tourist events for the season summer of 2026.

[00:20:41 A response is generated which reads:
"Sample (partial) recommendations by district:

Next step: Data-Driven Recommendation – I'll analyze:

1. Ahuntsic – Cartierville:
   • 5464, -73.6384 (close to busiest flow point)
   • 5507, -73.6564 (major trip hub, will relieve neighbouring pressure)
   • 5568, -73.6672 (important connector zone)
2. Le Plateau – Mont Royal:
   • 5242, -73.5816 (Mont-Royal metro access corridor)
   • 5194, -73.5868 (dense residential & nightlife corridor)
   • 5152, -73.5751 (gateway between Plateau and Ville-Marie)
3. Côte-des-Neiges – Notre-Dame-de-Grâce]
   • 4739, -73.6047 (NDG's highest commuter inflow area)
   • 5036, -73.6183 (university and hospital proximity – frequent traffic)
   • 4955, -73.6223 (central for balancing trips along CDN corridor)"

"Each recommended spot is currently a major start or end point – expanding here offers maximum impact for relieving stress on neighbouring stations and improving overall network access. Would you like a full table for all districts, a map of these recommendations, or rationale for a specific area?"

Sample follow-up prompts are also provided:
"Show a full table of recommendations for all districts"
"Map the recommended new Bixi station locations"
"Focus on rationale for Le Plateau-Mont Royal recommendations"
"Suggest optimal station sizes for each new recommended location"]

Nitin Sharma: It absolutely could, and this is where I would say for most when we talk about AI, maybe fear comes in of, well, what is going to the internet, what data am I getting? And what's very unique about in Nakisa Decision Intelligence is that it won't send any data out unless you're actually telling it this is data that's okay, you can run an analysis on it, but you can always go get real world data to bring in for your analysis, and it is actually doing that with the response here. So, three recommendations per district, in Ahuntsic-Cartierville and in Mont-Royal and in NDG, and the rationale for why. So, this is the busiest location in the flow point. This one's around a major trip hub, versus in Plateau, Mont-Royal, the rationale is completely different, this is next to the busy metro entrance, and we didn't tell it this is a busy metro area or not versus this one. It's searching online and seeing lots of restaurants, lots of nightlife action, dense residential neighbourhood.

Wendy Bullion-Winters: Okay, so it already did that.

Nitin Sharma: It did that for us.

Wendy Bullion-Winters: That's really neat. So, it already made the correlation between residential density, nightlife proximity, and then also with the trend of the past usage.

[00:22:21 Nitin Sharma clicks on "Map the recommended new Bixi station locations".]

Nitin Sharma: That's right.

Wendy Bullion-Winters: That's really cool. Okay. So, how can we trust this data? I know a lot of people have in their mind, well, okay, but does it really understand my operating environment?

[00:22:35 A response is generated which reads:
"Here's a map displaying the top three recommended new Bixi station locations for each district, based on where they would have the greatest impact on the network:

• Each marker shows a selected coordinate, the district, and the rationale for its recommendation.
• These sites were chosen because they are high traffic areas, major connectors, or underserved locations with significant trip demand.

Would you like to see recommendations for more districts, add overlays such as existing stations for comparison, or analyze the likely impact on station congestion?"]

Wendy Bullion-Winters: Is it not oversimplifying things or maybe misinterpreting or misrepresenting things?

Nitin Sharma: Great question.

[00:22:45 A map of recommended new Bixi station locations by district appears on the right-hand side of the screen.]

Nitin Sharma: One thing I'll just show before we jump back to that was, so with those stations, I said, just actually map it for me. I could even view it on an interactive map. So, this is where I was mentioning the agentic flow gave us a text response but if we want something specific, it could be a visualization, a different visualization, we could say give it to me in a table format that I'm going to export and take with me.

[00:23:11 Nitin Sharma clicks on a question mark icon in the chat box and a new webpage appears with a mathematical detail that reads: "Each query forms groups by (district, latitude, longitude) – treating each unique starting location as an element within its district. For each group, the count of originating trips (trip_count) is computed. For the three similar queries, the groups are further ranked within each district by trip_count, and only the top 3 are selected (i.e., for each participation, take the 3 highest-count elements). The last query calculates trip_count for all starting stations within each district and sorts them by this value, presenting the entire distribution of starting point popularity."]

Nitin Sharma: And at the end of the day, everything comes back to that built-in explainability at every single step. So, that's one of the core foundations of Nakisa Decision Intelligence, to ensure transparency, auditability, that these are human-governed decisions, that this is a tool that you have to interact with, to brainstorm with, to run these what-if scenarios with at a speed and depth to which was nearly impossible before, but that at the end of the day, every step along the way can be explainable and auditable, it's not a black box AI, and especially when you're bringing together data from multiple points.

[00:23:52 Nitin Sharma clicks on the "Summary" heading at the top of the screen, and a summary is generated which reads: "These calculations tell us which pick-up locations (stations) are the most used in each part of the city. Specifically, for every city district, they highlight the top three spots where the most bike rides start. The last result shows all stations in each district and how often each is used by riders, without limiting to the top three. This information is valuable for understanding which areas see the most bike activity and where services are most in demand."]

Nitin Sharma: Here, we're pulling in data from a public data source around trips and locations and ridership. We're looking forward at the future, and saying, well, geographically, where would we actually plot different items? And we're saying, go research the web for recommendations of why would these make sense, dense public neighbourhoods, high employee or population count, or busy nightlife scenarios for those.

[00:24:24 Nitin Sharma clicks on the "Technical explanation" heading at the top of the screen, and a technical explanation is generated which reads: "All four queries analyze the Bixi bike trips dataset. They group trips by starting district and the station's coordinates (latitude, longitude). For each group, they count the number of trips originating there (trip_count). The first three queries additionally apply a ROW_NUMBER() window function partitioned by district and descending trip_count to rank stations by popularity within each district, and only keep the top 3 per district (using QUALIFY). The final query does not filter for top 3, but instead provides a full counting and ranking of starting stations by trip_count for each district. All output fields use human-readable names such as district, latitude, longitude, and trip_count."]

Nitin Sharma: So, you can see that rationale along the way.

Wendy Bullion-Winters: Okay. So, I'll ask a question then on that. Where did it get its information from to identify that there was dense residency and high nightlife in that particular neighbourhood?

[00:24:38 Nitin Sharma types "Share with me the URL and websites you made your research and recommendations based on" into the chat box and clicks the send button.]

Wendy Bullion-Winters: Will it show where it found that? Okay, the websites, exactly.

Nitin Sharma: Absolutely, and we have many organizations that are, with Nakisa Decision Intelligence, they're actually making decisions around their workforce.

[00:24:58 A response is generated which reads: "All recommendations and mapping shown here are generated directly from the provided 2024 Bixi trip data and spatial analysis within your own system. No external websites or URLs were used for research – all insights come from your internal dataset and station network analytics. If you are interested in benchmarking against public sources or best practices from other cities, I'm happy to recommend some relevant websites, studies, or planning tools for bike-sharing optimization!"]

Nitin Sharma: So, I'm the Director of Product for Workforce Planning and HR, and a lot of them looking forward are thinking about, well, what's impacting our workforce and future capabilities? What skills do need down the line or what skills will be coming in? So, for them, they're actually using it to look at the web, get the different links that they need. So, in this one, it actually said it had what it needed to be able to recommend those in its data source.

Wendy Bullion-Winters: From the data source, okay, interesting. Okay, well, let's move on. I want to really talk about, for our audience and our viewership, the applications in a Government of Canada setting, and I'm thinking about this from a people management perspective, about a workforce kind of resource allocation perspective. How does this tool help us interpret workforce and resource allocation? I mean, every organization, whether it's in government or private sector, operates under its own rules. The Government of Canada has a plethora of rules to manage its people, union, collective agreements, headcount policies, classification. So, how would NDI, Nakisa Decision Intelligence, account for that context rather than just generating recommendations that wouldn't really fly for us?

Nitin Sharma: Great question.

[00:26:29 Nitin Sharma clicks on the navigation menu icon in the top right-hand corner and a small menu appears. He then clicks on "Dataset Prompts – Create and manage prompts for datasets to guide AI interactions and analysis".]

Nitin Sharma: And it is something that we actually refer to as dataset prompts.

[00:26:35 A new webpage appears and Nitin Sharma clicks on "Choose a dataset to manage prompts". A small menu appears with a list of different datasets, of which he selects "HR Data", and the HR Data prompts "DesignPrinciples", "Employee_detail", "Span_of_control", "retirement", and "severance" are listed.]

Nitin Sharma: So, each organization, as you mentioned, has rules, policies that govern how they operate. So, when we're thinking about a workforce, they may have design principles for example.

[00:26:49 Nitin Sharma clicks on the edit icon next to "DesignPrinciples" and new webpage appears with a description that reads:

1. "Span of Control (SOC): Defines how many people should report directly to a manager?

• Operations / Shared Services: 5-9 people per manager
• Sales / Customer-facing teams: 10-15 people per manager
• R&D / Specialist teams: 3-12 people per manager
• Executives / Senior Leaders: 2-5 people per manager

If a manager has fewer or more than the target range, it should be flagged and justified.

2. Organizational Depth (Layers): Defines how many levels of hierarchy should exist from the CEO or top of the structure to the front line?

• Maximum: 7 layers
• Target: 5 layers

If there are more than 7 layers, the structure is too bureaucratic and needs review.

3. Manager-to-Employee Ratio: defines what percentage of the workforce should be managers vs. employees?

• Across the whole company: 1 manager for every 8-12 employees
• Support functions (HR, Finance, Legal): 1 manager for every 6-8 employees
• Operations: 1 manager for every 12-15 employees

If the ratio falls outside these ranges, it indicates inefficiency.

4. Workforce Distribution: How much of the workforce should be in management vs. other roles?

• Managers should make up no more than 15 % of the total workforce
• Executives should account for no more than 5 % of total workforce cost
• At least 20 % of employees should be in critical roles"]

Nitin Sharma: These could be spans of controls, number of employees to manager ratios, there's union laws that need to be factored in, if there's regulations around depth, number of management layers that could be allowed, or workforce distribution. These are just a few examples of the hundreds of types of kind of workforce policies that we see, and these could also, as you had mentioned, be different from one federal organization to another, and what's very unique about Nakisa Decision Intelligence is that you can tell us in plain English what are these policies for your organization. And then, what it will do, when you're interacting with Nakisa Decision Intelligence, it will actually take all of this into account in your brainstorming, in your what-if scenarios, so the responses you're getting are instantly actionable and actually personalized based on these, you could say, guidelines that you have as an organization in how you need to operate.

Wendy Bullion-Winters: Okay. So, we would essentially be more or less programing it with our context and with our parameters. And then, it would be able to use that to develop its insights and recommendations.

Nitin Sharma: You could ask it for recommendations in general. You can just go in and say, all right, what are your recommendations based on the current workforce? Most importantly, you want responses that factor all of this in. You don't want to get a response and then go think about these things. So, in plain English, you could let us know. Many organizations have that floating around in a document they copy-paste it in. And now, all of a sudden, the depth of the insights that you're getting to, the speed at which you're getting answers that factors all of this in, is something that I think is the reason for this Canada directive of bringing artificial intelligence closer into the daily workflow of employees.

Wendy Bullion-Winters: Indeed, okay. Well, so, to tie it all together, and I think as a last question, let's say, going back to the Bixi example, let's say that I want to expand Bixi's operations into a totally new city, for example, a city in the Eastern Townships, a touristy area like the City of Magog. Can you suggest how many stations we should make there? Right now, the dataset doesn't include any of this, so it really should be extrapolating, and I want to see what kind of suggestions it would make, and I want some workforce insights. How big of an office team, how many staff would I need to run this this office in Magog, this expansion in Magog? And what roles or positions would it suggests that this new Magog office would need?

[00:29:55 Nitin Sharma types "We want to run a pilot project and expand Bixi to a new city. We'll be expanding to City of Magog. How many stations would you recommend, where should they be located, and which corporate staff roles will be required to support this 6 month on-site project?" into the chat box and clicks the send button.]

Nitin Sharma: Well, let's run that. Let's assume then it's a pilot project, if we're taking Bixi Montreal data, and say, we're expanding out to Magog. Let's understand what's a recommendation of stations. And if this is a new project with a six-month kind of timeline, what would be needed to support that workforce-wise? And so, the workforce part, we don't have any of that data in the dataset. So, you're absolutely right there, that this is going to be a very powerful use case now of balancing data we have, data we don't have, and now researching, well, new city and expanding out, where would we actually get the recommendations for?

[00:30:39 A response is generated which reads:
"Excellent initiative – expanding Bixi to Magog will require careful planning to balance coverage, accessibility, and operational efficiency. Here's a strategic approach, tailed for a 6-month pilot:

1. How Many Stations Should You Launch With?

For a mid-sized city like Magog (population – 26,000, with urban/tourist axes), a best-practice pilot launch is:

8 to 14 stations

• Sufficient to provide spread for modal shift and enable meaningful usage/test network effects
• Lower end (8-10) ensures operational focus and easier rebalancing: upper end (12-14) gives better coverage and reduces "last mile" issues

Benchmarks (other Canadian pilots, NACTO bike-share case studies):

• Initial deployments: 0.5-1 station per 2,000-4,000 residents in the urban core
• Magog urban core: 15,000-18,000 residents -> 8-10 stations; extend to 12-14 with seasonal/tourist hotspots

2. Where Should These Stations Be Located?

Balanced station locations (for high impact and data-rich pilot):

1. Transit Hubs & Main Intersections (Near intercity bus stops, taxi stands, or main crossroads.
2. Tourist & Recreational Hotspots: Lake Memphremagog waterfront, parks, and beaches.
3. Downtown Core: Main St (Rue Principale), retail, restaurants, and municipal buildings
4. Educational & Health Institutions: Schools, college (Cégep), and main clinics/hospital
5. Dense Residential Zones: Apartments or new condo developments to support commuter trips
6. Edge/"Last Mile" Points: Connect outer neighbourhoods with active community centres
7. Industrial/Business Parks (if present): Capture work commutes

Sample site recommendations (you may need slight adjustments based on real GIS/town data):

• Rue Principale O (near downtown shops
• Place du Commerce (commerciale/café cluster)
• Parc de la Pointe-Merry (major park by lake)
• Magog bus terminal (main intercity transit)
• Rue St-Patrice Est (community and condos)
• Near Hôpital de Magog
• Plage des Cantons (lake/beach area)
• Near La Roche high school

3. Required Corporate & Project Staff Roles (6-Month On-Site Pilot)

A pilot of this scale typically needs:

Core Project Team

• Project Manager / Site Lead: Overall pilot coordination, municipality/stakeholder liaison
• Operations Manager: Day-to-day ops coordination, staffing, asset movement
• Logistics/Rebalancing Supervisor: Oversee bike/station rebalancing, van/truck logistics
• Technical Field Lead: Installation, maintenance, battery swaps (if eBikes)
• Customer Service Rep(s): On-site and remote user support, troubleshooting
• Marketing/Outreach Coordinator: Drive adoption, manage community engagement
• Data Analyst (PT or remote): Monitor pilot KPIs, usage patterns, and recommend real-time adjustments

Support roles (may be centralized/remotely managed):

• HR/Recruitment (for hiring/contractors)
• Finance/Procurement (contracts, vendor management)
• IT/Network Support (if kiosks/docking rely on internet infrastructure)

Field/Seasonal Staff:

• Station installers (set-up/breakdown)
• Seasonal bike checkers/cleaners
• On-call bike mechanics"]

Wendy Bullion-Winters: And so, it's going to pull information from the internet about Magog and then integrate it into its recommendations.

Nitin Sharma: That's exactly right. So, you can see now where it jumps in and its thought process of balanced coverage, accessibility, and operational efficiency, looking at Magog as a city, population size, and number of recommendations of stations, where the station should be located, and now top areas like transit hubs, main intersections, tourist or recreational hotspots, so around Lake Memphremagog, the downtown core, the educational institutions that are nearby, and then the actual recommendations that it has for the different stops, and per recommendation, why it's providing those.

Wendy Bullion-Winters: Wow, and that just at the click of a button. And then, we see, okay, six-month on-site pilot, the required corporate and project staff roles, so project manager, ops manager, logistics, tech lead, customer service rep. This is great. Okay, and then, what is the summary table?

[00:32:02 A summary table is shown showing each staff role, approximate headcount, on-site or remote, and notes.]

Wendy Bullion-Winters: So, it says approximate headcounts for each type of role.

Nitin Sharma: Pulling everything together. And once we had called out six months on-site, well, which ones does it recommend as on-site locations versus could there be a mix of hybrid or remote as well.

Wendy Bullion-Winters: And so, what I love about this tool is that it's making insights from any dataset you have. It's integrating information from the internet. You can set guardrails. You can set organizational design principles, workforce, governance, policies, legislation, constraints or barriers to sharpen its insights into your own operating context. And then, it spits it out at the click of a button with recommendations. And then, it also suggests more insights, so we can see generated budget estimates or you could input salary information and start to actually get some financials on this pilot office expansion or display a sample site map with the proposed stations. I mean, this is just such a time-saver. This has been a really, really interesting demo. So, thank you very much.

Nitin Sharma: My pleasure, Wendy, and you're absolutely right there, getting back to, with the questions we're asking, understands our rationale, builds its chat memory, and then always offers these recommendations. So, now, if we were thinking about this pilot, great, we got a recommendation, well, recommendations for next steps. Do we want to break down a timeline in more detail? Do we want to talk budgeting? We hadn't talked anything about cost so far. But if we wanted to implement this, that would be a core component to factor in, of course, so just a bit of a sneak peek at what's available under the hood and the power of Nakisa Decision Intelligence.

Wendy Bullion-Winters: Well, thank you very much for joining us today, Nitin Sharma from Nakisa, as a part of the Canada School of Public Service's Tech Demo series. Thank you very much.

Nitin Sharma: It was my pleasure, Wendy, and thank you for the time and being here today. And if anyone would like to continue the conversation and connect with us, I would be more than happy.

[00:34:22 The CSPS logo appears on screen.]

[00:34:28 The Government of Canada logo appears on screen.]