Hello. Hi, can you hear me? Yes, yes, are you Jie, yeah, I'm COVID. How are you Nice to meet you? By the way, pardon me,
nice to meet you. Can you hear me?
Yeah, yes, yes. It's not. Kind of the audio is not optimal. To be honest, you're kind of on and off. Maybe that's my problem. Okay,
yeah, I mean, I mean coffee shops. I'm not sure if this is the environment. The environment has a lot of noise, so I'm not sure if that affects
I think for now, it is okay, but if it gets worse, I will let you know.
Okay, sure. So, okay, yeah. How to restart
shortly? Introduction of myself. I'm a second year, if that's okay with you. Okay, sure, so, yeah, I'm second year PhD in MIT, and I'm in transportation department, which is where Jie is the director, and I'm interested in startup operations. And my advisor is Charlie fine, who gave a talk in AI and entrepreneurship. I wonder whether you were in the July symposium.
Yeah, I was there. And I know Charlie very loud because he stayed at my office for afternoon, so he borrowed my office and I was also in my office, so basically we share the web office for my afternoon. So that's why I know him very well.
I see he's such a nice guy, isn't he? Yeah,
he's He's very kind,
yeah. So I'm that guy's student.
Yeah, you're very lucky to have him as your supervisor. So, yeah, I can give my introduction after you, after you finish your
Oh, yeah, I'm almost done. But I guess related to today's kind of motivation for me to reach out to you was, first of all, I like the word Tom Malone and Abdullah was doing I took Abdullah's class last semester, which is not directly related to human LLM interaction, but more on there is like, Do you know his paper on beyond 20 questions?
Okay, I don't know. Jie,
but the highlight, yeah, is social science is not cumulative, so we need to make the theory development more systematic and testable, and that is not irrelevant with his work on human LLM, because there are some areas where human, The scholars, current scholars, are really bad at doing replication and reproduction of some research, so maybe that might give him some motivation to argue that some areas of the task space machine is actually better. Um, yeah. So long story short, I think his idea is very applicable in entrepreneurship theory and in practice as well. So I want to build some entrepreneur operation model that somehow is programmatic and can be used in practice. So that's where I'm heading.
So I guess, yeah, my name is Jie, and I'm currently a postdoc under Supervisor of Tom and so, yeah, so I'm based in Singapore, and you were there, right? You were in July symposium, so you are in Singapore at that time.
Have you met by any chance?
I'm not sure actually, he was there. I was also there, yeah,
well, maybe we may already met. Were you there in the Winter Symposium as well?
Winter? You know, because winter I, I, I was preparing my graduation for TP, so I didn't have had the time to attend the winter time, yeah, but maybe we have met sometime, but, but so we know each other from that. So, so, so, basically, my tasting work is about how to use AI or like language model to support those social science research methods like qualitative analysis. And I built a lot of applications like colab coder and other applications to support those tasks. But after I built more and more applications, I realized, I realized that human and AI when they work together, especially right language model, when human and right language model work together, the country, there is no theory to support the building of such type of impaction between human so that's why I was motivated to start this work With Tom to build, to build the theory between human to build a modes, the taxonomy of interaction modes between human action to support those AI developers and designers, popular developers, to when they, when they want to build a sub applications some area, they should have a systematic way to think about which dimension I should consider to build this application, For example, for example, how, how
you want to stage? Ai should be incorporated. And also, in
how many people can, can we incorporate editing and other questions like, How can we how to use, how to design those power the system, to support to the human, yeah, more generally, these questions and so, so this can organize a reference for both of people, and also this can support them to create a new create a new system, basically new, new interaction mode. So basically we we build this theoretical framework from those existing literature, but we hope people can build new interaction mode from their own practice, based overview, yeah, that's basically the work that you, you have heard in last meeting.
Yeah. So very interesting. And
I saw, I saw you are interested in this direction also. So I was, I was wondering, what, what? Maybe, yeah, maybe we can get more specifically. So maybe which part you are.
Maybe, first of all, I
can clarify more questions about that work, and then we can talk about some tasty ideas.
Yeah, so
just, just to let you know, the audio is kind of on and off, I'm sorry, but
I'm sorry because I'm maybe let me, yeah, I'm using my airport and yeah, hobby shop, so
maybe let me Turn off my airport and just use
Thank you. Yeah,
okay, can you hear me now?
Yeah, way, way, way, much better. I so So, so thank you. I
Yeah, so I put it in the agenda that i If you don't mind me just cutting to the chase or, like, jump into the topic. What I was most interested in was applying, as Charlie suggested during the all hands meeting, applying the established framework in t5 T6 and apply that to T sevens, gate allocation, or the C CAG. And I've been chatting with young Ling. You know the from CAG and I specifically asked whether the AV, the truck, the baggage handling, how that is being managed, and because for me, it would be more kind of for my personal kind of interest, I would be more interested in innovation management, not just using the existing measure without any new technology adoption, instead of doing solving the optimization problem with that, I would be more interested in as the new technology is adopted compared to like human versus human LLM is, and how to evaluate this more dynamically so that we can evolve that dynamic evolution is my interest, and how the evaluation measure should somehow adjust to justify the adoption of the AV and Yan Ling told me that for now, what they're doing is like we first measure the first and the last bag when the first and last bag touches the arrival carouser. This allows us to know when the passenger receives the bag. So she was trying to describe what they are currently doing. And I think the high level idea is, as long as the machine mimics the human behavior, the workers then it pass. It is a pass. But I feel from your research and your expertise, maybe that there might be some better way to measure how the human baggage handling workers and machine AV can be evaluated better. Does that make sense? Like my curiosity?
Oh, yeah, thanks for sharing this agenda. So what say? The full name of AV, easily, autonomous vehicle.
Oh, yeah, I can. So
in the chat airport industry news and they introduce Changi has been one of the first airport to introduce that. Yeah, autonomous,
correctly? I see so A
is autonomous? What is very badly,
I think vehicle, autonomous vehicle. So instead of humans driving on the like baggage, the customer baggage full of cart, now the machine is automatically driving, just like Tesla drives on the highway.
Let's see. So now, what is, what is your current, your, your your, your guys, current progress in in developing this taxonomy, pardon me, yeah, what, what is your current progress in developing the structure the taxonomy. So, yeah, 000, so, so, all right. So you mean you guys are still in the idea ideation stage right?
Oh, just to let you know I was in t4 and t5 and Charlie thinks that T6 and T7 is more kind of of his interest and his expertise, because he's operations management person, and he thinks the CAG and kind of part of the marine the allocation problem is very classic operations. So he has recommended, he recommended me to get interest in T6 and seven. So I'm learning,
yeah. So yeah, I'm not sure the full mechanism that he's interested can overlap with T6 but yeah, basically, T6 have a lot of we have several postdocs and and for me, my I'm only working on, I'm mostly working on collaboration with Synergy Marie instead of CG. So and also I'm working on the research projects in human interaction modes taxonomy. So I guess the most relevant and the most the most relevant overlapping would be the human interaction taxonomy. So I can, maybe I can share some my experience in building this taxonomy, and let's see how much so, so I can share the methodology we used, and then let's see all that can help you guys to build this taxonomy. Is that okay? Yeah,
yes. And during your talk, you were mentioning something along the line of now that you established a framework, you would receive some data from CAG and try to combine with it. And I am interested in how that's going.
In that parts we so I can let me share Some of my experience. So
wait a second. Thank you.
Yeah, okay, can you see me? Yes, yeah. So basically, we are collecting the the process data from, say, Jie Bucha those, yeah, basically, those data are collected by, actually, so she's a professional senior consultant working with T6 help us to collect data from say, and it doesn't she also helped us to import those data into here. And also so we are. We currently have this website, and this web application allow people to input things, input process data. So the process data is like act and create a modified and so these are the high level branches. And I think the sages data is under modified. So let's see modify, oh, modify physical object and from
Yeah, basically these are all the other category, all the categories, and this is handle, handling, bagging and so very Thank you. And so, basically, for every activity, we will have the description for the activity. And we So, for example, human carries back from trailer to bottom conveyor belt, and we have description the process of manage, managing passenger luggage. So this is the, basically, a more specific description for this activity. And also we have, we have the actors, so say, for example, baggage handler, auto, automatic, baggage system. So these are the people and other components to help to form this task and the generalization, the general generation is a more generalized task or activity for this specific activity, and we also have specialization and parts. So basically this part part is the most specific sub activities for this activity, and you can see our change location, package object to be moved, position, serve, near object to be moved, load, unload, remove packaging. And also have some pre condition, post condition. So before this before we importing this data, we use another template to collect this one we use
so if you for every specific activity, you have a template, and it basically have title, activity, type and actors, pre condition, post condition. The most important part is this one the process, so one specific activity can be split to multiple sub activities, and there are dependencies And the rules inside this whole process. So, yeah, for many activities, we collected information in different dimensions. The interesting points of collecting those different dimensions for different activities is we can see their over life. So say, for example, for some activities, you have, you have, you already have some sub process. This process can sometimes be applied into other activities also. So, yeah, do you have any questions?
So you're trying to say that if a can be decomposed to a, one, a, two, A, three, and B can be decomposed into b1, b2, b3, but a, one and b1 may be the same thing, and you can somehow pull this and do this at the same time.
That's yeah, yeah. Basically, that's correct. So, so, so there are several ways we are envisioning this process mapping can can do some innovation. One is so the most important part is inheritance. So say, for example, this, in the in those process, there are a lot of sub activities, and for those sub activities, you can always decompose them into sub sub activities. And for, for, for this dimension, like pre, pre conditions, post conditions and the evaluation dimensions, all of these dimensions can can be inherited by the Sub Sub Sub activities. So basically, the satellite activity, say, for example, let me think a specific example for it. So I
Okay, so, for example, the dimension actor in in this in this activity, secure security screening vectors are traveler, baggage handler, check checkers and for their sub activity. Once you decompose it, it, you can use another template to fill in the whole dimensions. And those dimensions mostly are similar to these dimensions. So and you can change some part of this, of these dimensions to create a new template, create a new data point for the sub activities. Is that clear? So let me have this weird
you're trying to say that even if the activity changes the structure of having pre and post and evaluation that stays the same.
Mostly are the same. So. So for, for example, for this child node of this one, they all of the dimensions are similar. All of the dimensions are similar, except some part of it, because it's, it's another equity, yeah. So this, this arrow represents inheritance, inheritance relationship.
Oh. Um, so handle baggage has three inheritance I mean, like it's inheriting three of them or only one of them.
All three of them are inheriting properties from this one. So do you call this sub activities? We can decompose handle baggage into these are actually the specializations. So they are not the sub activities for for each one, for for this one, it has, it's actually have different the different parts or some activities are different from their from their specializations.
And, yeah, this is a little bit confusing. I can use, I use another figure to show you.
Yeah, we had this meeting before, and the Charlie is there, and
yeah. So
in in this figure, you can see activity, selected job, candidate by democracy. And it has four different directions. The in the later part you go to the generation it is selecting job candidates and the right part is different specializations. You for example, you can select a job candidate by democracy with majority vote, or you can select a job candidate by democracy with Priority vote. And the different parts is the sub activities. I mentioned it to you. It's so if you want to perform this activity, it will have four different parts. The first is group members learn about the candidates and group members vote for candidates, yeah, and the group leader. Cons will. Group leader announces winning any candidate. So this is a different from parts and specialization. Yeah, I know it's very
I see there is a similar like Korean concept, punk and blue, that somehow maps with this. So,
yeah. So I was also very confusing initially, until I see this figure, yeah. So in the top it is the use so basically, what use cases it had for this specific activity, so it will have a higher it is new employee. It can, you can be used to hire new employee. Yeah. So
can I say that select job candidate by Democracy is part of hire new employee.
It's part of which one.
So let me draw it. So since this is like a, b and c, b is, C is part of B, and same reasoning, B is part of a, right? I?
So if you see this direction, yeah, if you see this direction, it's it's only it's only a single direction. It's only a single direction. So these parts are not the another part of this one, because it is this. This arrow is starting from this, from this B to this a.
So, so
you can think about the different nodes here. All of them are, got it at central node, and for each central node, for each central node, it will have four different directions. Then these directions are actually, actually these dimensions. So when one potential use case for us is, perhaps, say, for example, select a job candidate by democracy with majority votes, the if I, if we, if I want to envision the specific parts and specific use cases for this one, most of them, most of them, you can you can find it from its parent node. Now you can find it from selecting job candidate by Democracy,
so yeah, may I
ask how often that you have meeting with CAG? I'm not sure. Recently I'm not really meet with but, but I work with early this year, and after that, I mainly work with Synergy Marine. So it is another local partner, similar to synergy Marine. And I think if you are more if you are very interested into the process, data collection, you can have a chat with each way. I'm not sure you have heard of her.
You know how I can get in touch with her?
I guess.
Let me find the per email for you. You
but she I don't think she used slack occurs, so I guess it would be better to contact her through email
and just to make sure I remember her rule, Oh, thanks. She is the one who is collecting or the documenting the CAG actions, right? Yeah,
she, yeah, it's she, she, she's the one part in two six collects the data from say, Jie and she has done a lot of work in mapping the process data.
So compared
to synergy marine and CAG, I'm curious how much work of data collection is completed. I think t say Ji data CG has much, much more data than synergy, because for synergy, we are still in the stage of formulating the research questions. So they are interested in some apply AI solutions. And TCS are interested in the process might be and how human and AI can collaborate with each other. So we are still in the process. We try to have an idea about how to, how to combine these two interests, but still in the state of formulating the research questions? Yeah,
since I have much less information and knowledge about the domain there, if you were me interested in operations and innovation management, between CAG and synergy Marine, which project would you somehow be more interested in?
I, for me, I actually don't have a specific preference in terms of their in terms of the airport and the marine. The reason why I prefer to work with Synergy Marine, it's it's because we can get the data more quickly. So basically, once we have some question about the data. We can, we can directly send an email, and they will give, give us data very, very efficiently, but for stages, for stages one, we might need a little bit more time to get the data. So that's why, that's why we, I, yeah, after, after work with CAC for several months, and I started the work with Senator Marie, but, but I don't have a specific preference in terms of transportation. Free transportation, yeah,
no, yeah. I think Charlie would be really happy about your decision, because the paper that I just shared, it's about operations for entrepreneurs, and there is a case study of a startup that failed because they chose a supplier with a very long time delay. So he emphasizes the importance of having a very fast clock speed between the interaction. So I just captured the part where supply chain design of a fast interaction when your especially in early stage, is very important. And I guess that depends on how big the size of the institution you're collaborating with. I can imagine CAG is much bigger organization, but senior Marine is much smaller, so that for them, getting some help from M 3s would be more meaningful. So they're called very much. Is it the right? I
I'm not sure the size of the company, but CD is more. It's more Singapore government. It is regulated by Singapore government, so it has more regulations, but it's more like it's also global, but it's more private company. So once you you want, you want to data, and you don't have to follow a lot of regulations, they can immediately give you the data. So, yeah, so that just gave you a bit of context about that. But I do believe for cities, I'm not sure I feel like Siege is more famous all over the world than Sydney generally,
I tried to study a little about synergy Marine. And so Charlie's introduction was they are trying to implement a new procurement system, and that was where they want to adopt the AI. And they have 600 ships like a floating warehouse, and they over buy some spare parts, yeah, so describing this as an inventory management problem is that correct? They also
want to expand the expanded those 600 fleets to 1000 if to 1000 ships. In three years. So that's the that's where, long term goal, short term goal, yeah, so, so I did feel a lot of opportunities to we can get in this, in this going process, yeah, so 602,000
ships in three years. Did you see and
yes to
1000 Sorry, Sorry, go on first.
Yeah. That's why they are eager to see how human and AI can work together, because they translate. They rely on human leavers a lot so, so that's why they want to keep the current, current human resource size, but they want to go to 1000 1000 shapes in three years. That's a That's not easy.
Do you know whether
they have specific tasks to automate in mind?
Yeah. So there are a few potential ways to automate us specifically, we starting from the procurement process, and we we started once she starting a purchase request, how their procurement team to facilitate those purchase requests and to supply different goods. So the in this whole process, there are a lot of stages only, only relies on human, and it cannot be automated. So but, but they are. They're thinking about, what if we some for some types of for some types of goods, they don't need to, don't need a human to initiate those purchase requests inside the system will automatically to, automatically starting, start to send, sending those purchase requests, and automatically to get quotations and get orders from different vendors and get approval from technical superintendent. So for this whole procurement process, perhaps that there are some of them can be don't need a human to be involved, and there will not to have human arrows. So, so basically, this is some initial ideas. And they are also interested in how to say, for example, how to extract those expert knowledge into those, those and or other AI solutions, and then the new people joined. Once the new people join synergy, Marie, they will. How to say they will, they can utilize those empower the knowledge to to help those new people, because, because, because the training is very difficult and and very time consuming. Yeah. So this their, their initial, initial and I, I remember, Tom is more interested in mapping those process and mapping the whole procurement process and mapping the whole stakeholder, which is technical superintendents, they are Interested in to mapping the different process or this specific holder, and kind of see whether, whether we can optimize those process and to innovate the way they perform procurement. And this can be applied into many, many companies in many because every company need to have a procurement. Need to do procurement process. Yeah,
I see, I see, oh, does it somehow require some make and buy decision, meaning, Charlie has a paper on different steps to decide whether to automate the process or just what level of process, then you have to automate.
Sorry, I need to pick up a pick up call, and I will come back give one Minute. Okay, Okay, sure, sure.
Sorry, I come back,
so I just put in a chat Charlie's paper on make and buy decision, and it at least I just search, and there appears a lot of procurement keywords. So I just wanted to bring it on your radar, because, yeah, the make or buy decision appears a lot in operations, especially like whether to outsource or produce it yourself. I know the situation might be a little different, because the synergy marine would not have some plans to build a ship themselves, but when choosing the suppliers, I think the decision standard or tree can somehow be applied, if that makes sense.
Sorry, can you see again?
Yeah, sure. Hold on. So the paper. So my first reaction was, procurement appears in this paper five times, so maybe there might be some relevance. And I was trying to say that buy or make decision is fundamental, so let me just share the screen so we are seeing so some example is here, like, what is the history of development of infrastructure procurement patterns? So maybe that's one example. And so I guess the point I'm trying to make here is Charlie has some expertise in auto industries. And if you think of auto industries, you need to decide whether some parts of modules of the car, you need to procure from some external places if you don't intend to build everything in house. So,
yeah, he has some procurement process for each class of decomposability. Again, I have less knowledge about the ship, then even, even less than the car. And I don't know how the procurement process happens for each part of the ship or for the whole ship. I'm really excited to learn, but I'm just trying to make a point that there might be some existing make and by decision framework based on system engineering that we can build on.
Yeah, definitely. So there are already a lot of existing literature, existing process mapping, in procurement, in in by in buying activity. So, so So what would we what do we wanted to try is to mapping the procurement process with in synergy Marie, similar to to those. So, so we want to adapt to the current existing process mapping to their specific use case, and to see what are the specialization so basically, their use case is a specialization for our Whole ontology. And those, those, like I said, the dimensions, say by the actors, the evaluation, those dimensions for the literature, it can be used in their specific use case. So yeah, in the specific procurement process. And I again, I can, I can show, I can show graph. And you will, yeah, it's, it's a, it's a poster we presented in this July symposium. And,
ah, so it seems you have been working with synergy for more than four months, right?
I guess starting from early this year, because, as in the early this year, I, at that time, I worked with two industry partners, but I realized I didn't have so much time to work with two partners, and later on, I was only working with Synergy Marie. So as you can see, say, for example, this is the whole whole ontology, whole taxonomy of collective intelligence. And so it it is under this whole map. And I mean, I and there's a home map. There is a specific branch called exchange or modify ownership, and there is a specific activity, which is buy. And under buy, there are different parts or sub activities. And of course, it will have a lot of different dimensions like here, like the preconditioned post conditions and the evaluation dimensions, and all of those dimension shows will be inherited by this one, by the procure edge Company X. This company X is actually synergy Marine, and it will have identified suppliers, receive purchase request, select a supplier, and when under selected supplier, you will send a request for quotes to supplier, and you will need to analyze those requests for quotes to identify best supply. And then you place orders around deliveries and involves and manage supplier. So this, this actually is a more generalized it can be applied into multiple companies. And under this one, there is specifically signatures procurement process. So you can see there is received purchase request. And under this there are specific several sub activities and their Select Select supplier, it has different process. Sub activities like create request for quotes and send to a list of vendors request for quote emails from system and need to make decision, are those vendors willing to quote Yes, then vendors will reply with code. So If no, then they will decline email from the system. And then our purchaser need to find a new vendors to require support. And so this is the subactive is under the Select supplier. So, yeah, yeah, that's just a overview about how we how we map the procurement process of synergy. Marine incorporated this specific use case into our whole ontology.
So my question is, first of all, this is really helpful to understand the whole picture is kind of AI, are you? How do you test this? AI, do you? Are you trying to make an experiment that compares the manual work efficiency versus the AI efficiency by replacing it.
Yeah. So initially, initially, I was interested. I was interested into, say there is a specific person here, which is they call them purchase, execute executive. And at that time I was trying to, trying to, so the research question here is, can we, how can this purchaser axial tips? How can we extract our expert knowledge and in the whole process? Can we replace this person, because this person need to go through the whole process and monitor every specific steps and to ensure the vessels receive the foods, but But can we use those new, new people, people without a lot of experience to collaborate with light language model and collaborate with AI to perform this whole process. At that time, we will think about this specific reason question, but, but Tom? After we discussed with Tom, Tom is more interested into into another stakeholder, which is TSI. So if you check here it is once, once every process, process is is done, and they have, they need to okay here so they need to analyze the received code, and they need to save superintendents. And Tom is more interested into in the mapping the specific processes activity first, the processes for the for the activities this TSI is performing, for the everyday, for everyday task in the in the in the normal working Yeah, in the normal working schedule. Because beyond, for those TSI, this procurement process is not the main responsibility of responsibilities of TSI, but TSI have to spend around 30 or 40 percentage of his time to working on this specific procurement process. So that's why we were thinking, can we decrease the time this those tsis are spending in this process. Maybe we can, maybe we can automate some part of this process to decrease the time the TSI need to spend here. And we can, maybe we can have, there are two main directions in this time decreasing. One is, maybe we can utilize some AI alternative. Another is, perhaps we can optimize the the whole process. Yeah, we can optimize the whole process to decrease the time Jie size need to spend here. Maybe we can switch the process. Maybe some activities need to be performed before other activities. And maybe we can borrow some process from other domain, from other companies, say, for example, the procurement process in ECG, for example, you say, you say, they have used some interesting process, and we can borrow that process and apply that process into here and optimize the procurement process performed by TSI. So that's that, yeah, that's that's the two direction we are currently pursue, and we are in the process of collecting the collecting the process data for TSI and and see more detailed, see more specific research questions from there. Yeah,
interesting. So TSI is not part of synergy Marine, or is are they? Yeah, that
is part of the synergy Marine, but you can, you can think about them as very expert, uh, expert people. They they are, they are very senior people, so they have to handle a lot of tasks in synergy Marine and the procurement process is just only one part of the job, but it takes a lot of time. So, so So there the company want to because, so, as I mentioned to you, want to increase the size of the company. They want to handle one ships in the future. So that's why they want to improve the capability of TI guys.
Yeah, so I So, just to confirm that I followed you, I think Tom thinks perhaps the TSI is the bottleneck, because they're busy and this process, testing approval time is very long, so by making it more efficient, they can fasten the flow of this whole process, thereby increasing the ship in a very short period of time. Yeah,
yes, yeah, that that's that's exactly what we are thinking about. And yeah, hopefully we can find some after, after we get all the data, we can find those opportunities to open it, optimize this one. This, yeah,
wow. This is, I think, much interesting than CAG, if I might share my screen really quick for one or 30 seconds, I was trying to make a kind of a simulation model for the CAG, but now I think maybe I should make it for the senior Marine, because that seems more interesting. But like what this was trying to do was Charlie shared the slide on the description of how the dynamic allocation works. And I try to I think your version is much better, but this is what I could do with the slide. Like there are some entities, I think, actors, in your land, taxonomy and some attributes and operations, events, constraints and probabilistic elements. I'm mostly interested in probabilistic elements because my background is probabilistic program. So like, for instance, rare event estimation or some expected utility calculation, that is what I can do well, so I wonder. I'm just curious whether procurement investment or the procurement operation includes some rare events, like tsunami is coming, yeah, but that aside, yeah, after doing this, what this dashboard is trying to achieve is their utility as given was weighted average of departure time urgency and passenger connection and crew duty limits, I think like 30% 40% 30% somewhat, and based on how you change that weight, you can somehow play around different weight. So the distance is how much um, kind of the the closer you get, you are somehow weighing this much higher. And there were some different examples of flight a, flight B and flight C, with, for instance, flight A is full of customers, and flight B is emergency medicine, so the criticality is very high. So this was trying to give some interactive tool that humans, those without some domain knowledge, who hasn't worked in airport for more than 20 years, can somehow learn how to decide in a much faster way. So
that's interesting.
We do this a lot in system dynamics, where we develop a dynamic model and play allow the user to play interactive game. So and I heard Tom Malone's thesis had some component of how to make learning faster, and one was fantasy and the other was challenge. And perhaps I may be remembering wrong, but one of them, I think, was a game. So yeah,
yeah, that's it. Yeah, that's interesting. So I was so you build this. This is this one? Yeah, can I interact with it? Can you share the link with me so I can sure?
Yeah, oh yeah.
The purpose is to is, the purpose is to help the decision makers to decide which flight should be the priority.
Yes, yes, yes. And if I develop more, I could somehow connect with some language model and give the reasoning why in this situation of the weight, why B or C is most prioritized, yeah, and I think this can be applied to procurement as well, because if You have three options and your decision standard somewhat may change, depending on, for instance, the cost of some parts. So maybe, let's say there's a safety condition and the cost condition and some carbon emission conditions, and depending on politics and the price changes, your weight for the decision would change, and depending on that, your choice of the procurement would also be affected, right?
Yeah, and,
but one, one question I have is eight this. Let's think about this one is applied to procurement process, but most of the steps for procurement a are fixed, so they are just those purchasers to perform the purchase activities follow those steps and but they do have some emergency things need to tackle. So say, for example, some orders need to be prioritized. Some some orders you can it's more regular. Is a regular supply, and some others are, I would say it's less priorities. And I, I, I, yeah, I would say there could be some opportunities here to apply this, this one. But I also, I was also wondering, so what is their current strategies to decide the priority for flight? So are they making those decisions just by human, by Yeah, just by human brain, just they don't have any to support this decision making. Or, yeah,
I guess since T7 is working on it very hard, maybe they'll gonna be automated soon.
So you mean, she said that is working on some tourists to help them to
so for Yeah, for now, I think they do have some dynamic prioritization matrix, but I don't think I haven't been, I haven't been interacting with CAG at all so, but I don't think they have some automated, dynamic optimization process that processes in stream data, determine the weight, and based on that, make decision. They don't have the infrastructure.
I don't think I see, I see,
yeah, I think so. I think you're, yeah. I think your your current, your current model, or your current this, I don't know how to call it this dashboard, but I can, can you add into Council, we applied, into the into the procurement process, and I, I do feel that I do feel that if the because procurement process occurs in every company, so if you, if your model is applied into here, you can, it will have a more general audience. It have, it'll have a lot even, even if you go to other other companies, even as a country, it will be similar in Jason, yeah, so that's, that's
one last question and one last comment regarding your comment about the procurement process is very kind of fixed at this point from your analysis, I think there is some rooms for improvement by considering different options in parallel. This idea comes from when you are looking for the investors, for the founders, mostly what people used to do was just pick with one investor at a time, but I am taking entrepreneurial finance class, and the lawyer was suggesting, why are you doing that? Just talk with three or four investors at the same time. That way you can somehow hedge some risk and know more about the kind of your value. So kind of applying this idea, maybe I don't know the reality in synergy. Marine group, can you do some two procurement arrangement in parallel and somehow choose one in the end?
Yeah, that's, that's an interesting idea. So we never think about, yeah. We never think about this direction, maybe, maybe once we have, once we have new process, innovative, we can apply this to procurement, to two procurement process in the same company, and to see which one, yeah,
oh, yeah, that would be really interesting, yeah. And so one last question, that was a con. One last question was, if I were to make this model summary for the synergy Marine, like, who should I talk with, and who should I kind of get some permission to get the data at least?
I guess, I guess maybe Charlie. I'm not sure if Charlie is involved officially in M 3s project, and I think you can, maybe you can talk to Charlie and Tom, and also, once we approve, and I guess you can work with Kali, I'm not sure if you know kahali. And you can, you can, yeah, to let her to sign on a form and just get the data permission, and then you can work with Synergy Marie data. So I yeah, that's this process. Yeah, let me know if you have any further question regarding the synergy Marie's procurement process. I Yeah,
