Neuro-Symbolic Agents in Action

04:58:07.330 –> 04:58:10.360
Julia Nimchinski: Welcome to the show. Tilda Ross.

1536
04:58:10.360 –> 04:58:11.140
Tooba Durraze: Hi.

1537
04:58:11.570 –> 04:58:12.720
Julia Nimchinski: What a pleasure!

1538
04:58:13.390 –> 04:58:15.120
Tooba Durraze: Yeah, thanks so much for having me.

1539
04:58:15.570 –> 04:58:18.509
Julia Nimchinski: I know you preferred something really special.

1540
04:58:19.420 –> 04:58:20.360
Julia Nimchinski: Hi des.

1541
04:58:20.360 –> 04:58:29.700
Tooba Durraze: We’re hot off the press. We’ve been not around for as long, and we actually are going to be demoing something we just released, I think, 3 days ago, now

1542
04:58:30.090 –> 04:58:32.109
Tooba Durraze: excited to share with everyone.

1543
04:58:32.280 –> 04:58:33.119
Julia Nimchinski: Can wait.

1544
04:58:35.090 –> 04:58:35.850
Tooba Durraze: Okay.

1545
04:58:38.910 –> 04:58:51.330
Tooba Durraze: awesome. So for everyone who doesn’t know me, everyone in the room. My name is Tuba. I’m the founder, and CEO of amoeba amoeba is your AI powered data scientist for go to market teams.

1546
04:58:51.640 –> 04:59:16.179
Tooba Durraze: We are neuro symbolic in nature. So I’m going to dive. I’m going to jump back and forth between slides and a demo, so to lend some concepts. So please bear with me on that. The thing that I was talking about with Julia just now is like we just unleashed pipeline intelligence which is powered by Amoeba’s agents. And now I guess I should say agency of agents, as they came up in our panel yesterday.

1547
04:59:16.180 –> 04:59:24.410
Tooba Durraze: So what is the problem that we’re trying to solve? Essentially, the problem that we’re facing today is you have so many tools that you use

1548
04:59:24.744 –> 04:59:29.429
Tooba Durraze: in order to drive your business or drive your pipeline in this case. But

1549
04:59:29.660 –> 04:59:45.429
Tooba Durraze: a lot of the tools focus on either the orchestration side or the amalgamation side. So the piece in the middle to figure out what’s working what’s not working, but figuring that out before it actually happens, there’s a lot that doesn’t exist in that space.

1550
04:59:45.640 –> 05:00:01.789
Tooba Durraze: So we’ve created something that is akin to how engineers take observability. So engineers, obviously, technical teams have a ton of observability tools that they use to figure out that something’s slowing down. Something’s not tracking well, because their whole job is to figure out

1551
05:00:01.900 –> 05:00:04.610
Tooba Durraze: the crash before the crash actually happens.

1552
05:00:04.650 –> 05:00:33.439
Tooba Durraze: So, taking that concept, we said, why would something like that not exist for marketers? Again, the problem that we’re talking about here is, you want to be able to sleep well, rest, and not have the cognitive load of like. What might I be missing in my data? Or is my quota tracking accordingly. You don’t want to end up in a C-suite conversation at the end of the quarter and say, like, Hey, I didn’t meet my goals. So we’re in the space of proactive decision making in that sense.

1553
05:00:34.400 –> 05:00:35.120
Tooba Durraze: So

1554
05:00:35.790 –> 05:00:58.410
Tooba Durraze: with pipeline intelligence. What we’re trying to do is it’s a unified pipeline observability tool. We give you proactive intelligence. And again, no heavy lifting. I know there were a lot of questions throughout this summit around like data quality, and you know how hard it is to maintain data quality or what kind of data. So we essentially take care of all of that at amoeba.

1555
05:00:58.410 –> 05:01:16.549
Tooba Durraze: And again. Why, we’re a little bit different than some of the agents that you’ve seen is, we are neurosymbolic agents in nature. So I will jump into kind of what that is for you in a moment. As I demo things. So you’ll understand, kind of the difference between neurosymbolic models versus transformer models which are like Llms that you see.

1556
05:01:16.550 –> 05:01:33.259
Tooba Durraze: So again, 2 principles that we abide by, it’s insights that you can trust. And second, it’s always going to be explainable. That’s a big part of conscious AI product design which is making sure that you have trust in the outputs that the AI system is putting forward.

1557
05:01:33.320 –> 05:01:37.740
Tooba Durraze: So without further ado. I’m actually going to jump into the platform a little bit

1558
05:01:38.720 –> 05:02:02.419
Tooba Durraze: so welcome to amoeba. We are right now in nucleus, as you’ll realize, because it is a data science platform and called amoeba. Things are named kind of in a very geeky sciencey way. And that’s intentional. So if you think about nucleus, you think about kind of like a central hub, and that’s what this place is. And this is where you would come to track all of your goals. So

05:02:02.830 –> 05:02:07.750
Tooba Durraze: nothing like anxiety of demoing something. Live. But I’ll go ahead and start my first.st

1560
05:02:07.760 –> 05:02:37.099
Tooba Durraze: Add a goal here, so your goals could be anything again we talked about. This is around tracking pipeline. But that doesn’t mean that you can’t have goals around like retention, as we were speaking about earlier. Can’t have goals around like growth. Cro related goals. Cfo related goals again. You can have all of those as goals, because they all have some sort of implication on how you’re driving kind of top of funnel business. But the intention here is like this is meant for pipeline generation. So let’s say, here.

1561
05:02:38.172 –> 05:02:42.400
Tooba Durraze: I want to grow my roas

1562
05:02:42.510 –> 05:03:06.249
Tooba Durraze: as an example, it’s a goal most marketers deal with. So again, for the purposes of this demo, we kind of restricted the data sources because we want the things to move a little bit faster. But in terms of data sources, you can imagine we are dealing with anything that exists within your normal realm. Right? So you’re thinking about like your crms, your your advertising data, etc, like all of that, can flow into this.

1563
05:03:06.679 –> 05:03:23.150
Tooba Durraze: So 1st and foremost one of the things that we do. We have our 1st kind of interface agent, which is Atlas and Atlas. Like Atlas. The name suggests its job is to kind of guide you into understanding what your goal really is. Now, this agent is a lot smarter behind the scenes than it would seem, because

1564
05:03:23.150 –> 05:03:45.559
Tooba Durraze: its job is to also anchor you into what might be the right goals. Right? So sometimes we get anchored to looking at goals that are too menial or too micro. This Atlas’s job is to make sure that you’re focusing on macro goals that grow your business, whereas there’s like micro goals that might not be as relevant to your business. So in this case I’ll come here and I’ll say

1565
05:03:45.990 –> 05:04:06.460
Tooba Durraze: I want my let’s say day 0, Ros to be 75. So again, you know, it’s a pretty high number. I think one of the things I joke about with our customers is like set, really, really ambitious goals, because you want this to be the system. That’s kind of looking at all of those goals, and, you know, like taking a look at your entire system that way.

1566
05:04:06.460 –> 05:04:19.930
Tooba Durraze: So again in this demo, because the data is really limited, it came back really fast, and it set the goal for 0 raws to be 75%. And then that’s the only thing that you’re doing so far. So I press. Accept.

1567
05:04:19.930 –> 05:04:45.520
Tooba Durraze: Now, while that’s loading, let me take a step back and kind of show you about what’s happening behind the scenes. So again, as I mentioned, we’re taking in data from all sorts of sources. The agent that you didn’t see that’s behind the scenes is helix, which is our transformation agent. So again, we’re connecting into your data sources as is meaning, you don’t have to take care of any of the data transformation data cleaning or any of that. Like a good data scientist, this platform will do that for you

1568
05:04:45.520 –> 05:04:56.800
Tooba Durraze: and atlas is something that you just saw. And this is nucleus is our neurosymbolic system. And again, amoeba, it’s an agency of agents. But that’s the platform that you’re experiencing right now.

1569
05:04:57.415 –> 05:05:03.410
Tooba Durraze: So I’ll jump back. So again our goal has been set, and we are in our 1st living brief.

1570
05:05:03.842 –> 05:05:13.779
Tooba Durraze: So here you’ll see you have your goal. So again, our goal is 75%. Our current average for this is 50%. This is basic math, not a lot going on here.

1571
05:05:13.780 –> 05:05:37.740
Tooba Durraze: But I’ll draw your attention to this portion. And actually, any lime green bubbles that you see is our system, basically explaining what’s happening behind the scenes, or how the data is derived. So in this case, it’s already starting to anticipate that there’s going to be a trend related thing that’s going to happen. So say, day in, day out, you can come back to the platform and look at how this brief is changing.

1572
05:05:37.740 –> 05:05:46.129
Tooba Durraze: or, better yet, we actually have digest. So we can send that brief directly to your email. So you can wake up and have peace of mind that something’s happening and happening. Well.

1573
05:05:46.380 –> 05:06:02.240
Tooba Durraze: So now, when I go down here, these insights are actually key drivers. They are both linear and nonlinear in connection in relationship, which means like they might be directly connected to your goal, or they might be multivariate and connected to your goal through some other channels. Right?

1574
05:06:02.240 –> 05:06:25.200
Tooba Durraze: So again, the job here is to tell you these are your key drivers. So you understand what’s actually driving your priority goal. And then for each of these key drivers, we then also give you a bit of an again an explanation of like, why, that is a key driver that is important to you. If you want to build trust in like what you’re selecting here. Now bear with us. Because again, the data source here is like really really limited. It’s like, not for the purposes of the demo.

1575
05:06:25.200 –> 05:06:33.360
Tooba Durraze: So some of this might appear to be like a little bit generic to you. But the more data that you connect in, obviously, the more you’ll find here. So

1576
05:06:33.360 –> 05:06:54.370
Tooba Durraze: now that you have insights. So you know what things are driving your business, the next step would be okay. You’ve given me this information. But what do I do with it? What do I do with this information? So then our system takes it a step further, and for each of these insights, we actually give you recommendations. So if you flip through these insights, you’ll see recommendations and one quick point here we

1577
05:06:54.410 –> 05:07:23.119
Tooba Durraze: kind of restrict these 2 from 3 to 5. And that is intentional in nature, because one of the things that humans generally have a hard problem with is focusing on the things that really actually matter when your data sets are so massive. So we focus on key drivers, there’s a weights and biases method behind the scenes that are really really pertinent to this goal that you have set. And again, like I said, the whole goal here is to not overwhelm you with information, give you the right information at the right time.

1578
05:07:23.510 –> 05:07:44.649
Tooba Durraze: So if I go down here and I look at some of these recommendations now, say that you’re now translating. You’re going from actual insight into some sort of an action. So I look at these recommendations, and you know some of the recommendations. Again, they’re multivariate in nature. So this they kind of span across your department, even though the goal across your organization, even though the goal was your department related.

1579
05:07:44.680 –> 05:07:53.210
Tooba Durraze: So in this case, say, like, I want to pick optimized pricing models. I would go ahead and select that scenario. I would go ahead and select that. So

1580
05:07:53.210 –> 05:08:21.570
Tooba Durraze: for each of those recommendations. We’re actually giving you 3 ways to implement that scenario. Right? These are 3 like simulated scenarios that you would take. And the intention behind this is again, this is a pick, your own adventure type of product, because where humans are really great is intuition. So we want you to be the person who picks like, how do I want to act on this? But we want to give you all the information and all the pathways that you can take to address it. And again, there’s a forecasted lift number here, and

1581
05:08:21.570 –> 05:08:27.049
Tooba Durraze: that’s really useful when you’re thinking about how much effort am I going to spend on this. And what is that going to do for me?

1582
05:08:27.090 –> 05:08:45.789
Tooba Durraze: So in this case, let’s say, I want to pick seasonal pricing strategy. Now, again, this is a little bit high level. I want to be able to translate that into a work plan for my team, so I will come over here and I will generate tasks. Now, what’s happening behind the scenes is we’re pulling in a bunch of different data

1583
05:08:46.270 –> 05:09:05.640
Tooba Durraze: to figure out like what will be the constraints in relationship to this scenario that you have picked. So in this case, it’s asking like a breakdown of what to evaluate. So I will say, Okay, let’s evaluate in some cases, if the data is like really, really not expensive enough, it will

1584
05:09:06.030 –> 05:09:23.100
Tooba Durraze: also be able to tell you that you are if the data is not expensive. I’ll also be able to tell you like this. This scenario is too limited right now, so you shouldn’t actually act on it, etc. So let’s say I want a budget of

1585
05:09:23.950 –> 05:09:25.690
Tooba Durraze: 5,000,

1586
05:09:27.320 –> 05:09:45.450
Tooba Durraze: so I’m giving it constraint. Its job again is also to guide you. If you don’t have the right constraint, and then, based on the constraints that you’re you’re putting in there. Look at that! It’s already dividing up the budget. It’s 1 constraint that I gave it. It’s already dividing up that budget to make sure that you are best suited to implement this.

1587
05:09:45.450 –> 05:09:58.809
Tooba Durraze: Now, if I click, accept, I end up in our tasks panel, which means we’re taking now all the way from the goal that you have set all the way in the actions panel. Now, how you would go about implementing that

1588
05:09:58.810 –> 05:10:22.860
Tooba Durraze: this. So this is a basically a project plan for you. You can take this project plan and you can push it to slack. You don’t have to do it in our system, but the idea is, everything is mapped out so you can just pass this over to your team, and then in some cases you may come here and you may say, Hey, I just remembered, based on this. I need to make sure that I have a web dev that I have registered. So I can obviously put in custom tasks

1589
05:10:22.860 –> 05:10:31.939
Tooba Durraze: like, I want to schedule a web dev because anytime you’re doing new pricing strategies. You obviously want to change around your page, and I submit that, and that gets added right here.

1590
05:10:31.940 –> 05:10:49.670
Tooba Durraze: Now let me come back and draw your attention to one other thing which is with all data, there’s always this idea of exploration of data. And we also really fundamentally believe that there are things unrelated to your goal that exist in your data that might be interesting to you.

1591
05:10:49.670 –> 05:11:07.690
Tooba Durraze: So if you look at the insights here, there’s a little toggle that says, view and data lab. And the idea behind the data lab is, you’re taking the data that was connected. And then you want to explore some weirdness or some trends around it, so you can come here and you can ask it a basic question like, what trends might I be

1592
05:11:07.750 –> 05:11:10.860
Tooba Durraze: missing, or give me some

1593
05:11:11.070 –> 05:11:23.589
Tooba Durraze: odd trends again. The idea there is, like most folks in marketing are really really good at now having experimental mindset while they’re running their business. So we wanted to give you both of those dimensions.

1594
05:11:23.710 –> 05:11:44.560
Tooba Durraze: and while that pulls up, let me point your attention to one more thing, as always with AI context is king. So we have the cortex here, and the cortex is there for you to give the system some context around how you run your business. Specifically. Again, this is an Internet facing tool, right? So we’re not looking at tonality, etc. But we’re looking to give it context around, like how

1595
05:11:44.560 –> 05:12:01.680
Tooba Durraze: you want to run your business so cheekily here. This is obviously Pied Piper from Silicon Valley is the company that I’m using, but the idea there is you can fill this as expensively or in as limited of a way as you would like. Now let’s go back here and then see

1596
05:12:01.860 –> 05:12:11.120
Tooba Durraze: we’re still populating an answer. But let’s see. So it basically is picking up trends. One thing you’ll notice, even though interface looks very common to transformer models.

1597
05:12:11.200 –> 05:12:23.929
Tooba Durraze: A neurosymbolic model only puts out long form analyses right? So its job is to find, like the really intricate details, and then not just give you that. But you can also again ask it, for, like what sort of limitations exist in the data.

1598
05:12:23.930 –> 05:12:45.250
Tooba Durraze: and the job of the system will be to guide you. So your day to day hub, is this, again, this nucleus, this cockpit that’s guiding you in the right direction. You have updates here that change as your data changes. But more importantly, you have this like living organism, living, breathing organism at your disposal to make sure your business is running as you would have wanted it to run.

1599
05:12:48.900 –> 05:12:49.790
Tooba Durraze: Thank you.

1600
05:12:50.170 –> 05:12:54.759
Julia Nimchinski: It was love, what you’re building. I think it’s 1 of the most exciting.

1601
05:12:55.050 –> 05:13:01.409
Julia Nimchinski: I don’t know. Platform technologies I’ve seen this year so super excited.

1602
05:13:01.700 –> 05:13:04.363
Tooba Durraze: I’m biased. So I say, yes, it is great.

1603
05:13:04.660 –> 05:13:10.119
Julia Nimchinski: It definitely is. And yeah, for everybody’s watching. What’s the best next step.

1604
05:13:10.750 –> 05:13:35.070
Tooba Durraze: Yeah. So come, reach out to us, reach out to me on Linkedin. You can find me or come to Amieveai and ask for a demo again. We just launched this a couple of days ago, and we’re super excited to be showing it to you. The capabilities of the Middleware go a lot beyond what we can fit into our ui so far. So if you have an idea around something else that you might want to do with

1605
05:13:35.070 –> 05:13:43.980
Tooba Durraze: a system like that. I’m also very, very happy to hear that as well. So, looking forward to showing this to you within your data as well.

1606
05:13:44.770 –> 05:13:45.810
Julia Nimchinski: Thank you so much.

1607
05:13:46.150 –> 05:13:46.920
Tooba Durraze: Thank you.