← Lead With AI
Lead With AI
Female Founder Turns a Million-Dollar Data Entry Problem Into a Simple AI Solution
Female Founder Turns a Million-Dollar Data Entry Problem Into a Simple AI Solution
Loading summary
Transcript25 lines
- [00:00]
A
In every company, there's a hidden monster. It's called paperwork. The endless stream of forms, receipts, and spreadsheets that drain time, money, and patience. But what if data entry didn't have to feel like chaos? What if it could run itself? Today on Lead with AI, I talk with Normeen Cosmi, who is the founder and project manager at Datrix. It is an AI power system that takes the manual out of data entry. It doesn't just process information, it understands it. From messy inboxes to half finished spreadsheets, Datrix brings clarity where confusion used to live. Let's get into it. Welcome to lead with AI. I'm Dr. Tamara Nall. In each episode, we will take you behind the scenes with visionary leaders shaping the the future of AI across public and private sectors. Join us as we explore groundbreaking projects and innovations that are transforming industries and making a real impact on people's lives. Let's dive in. So, hello everyone. How are you? As you know, I'm Dr. T, your host for Leigh with AI and I always like to start by saying thank you, thank you, thank you to all our listeners who tune in every week. As you know, we hit number one in technology on Apple podcast this year and we also are a gold winner for the W3 awards for our podcast. And I'm so excited and I always want to just show our appreciation for all of you who listen all over the world. And it also would not be possible with our great guests such as the one that we have today. So today we have Narmine Cosmi, who is the founder and project manager at Daytrix. Narmine, how are you?
- [02:02]
B
Hey, I'm really good. How are you?
- [02:04]
A
I am doing well and excited. I just love finding out about all these AI products, particularly when there is a female founder. So welcome to Lee with AI.
- [02:15]
B
Thank you so much for having me here, Tamara. I'm really happy.
- [02:21]
A
Okay, so let's start with who you are. Who are you at your core, what are you passionate about? And what was the moment where you decided that day tricks was needed?
- [02:34]
B
Okay, so it comes with like a really interesting background story. So I have been working in tech for almost three years now. And initially I was an AI developer. So I worked with like initially code. And then because I'm an extrovert at heart, I started, you know, doing some client management and project management. So I became like a solution architect. Architect situation where I talk to clients and developers at the same time. Right. And like a year ago I was working with this really huge textile client, right he has clients that are like, pretty little thing and boohoo. And we were making this AI automation for merchandiser for them. And the huge chunk of our time was going after, you know, fetching data and asking them for data, having clean data. We need this data somewhere else. We need it organized. The time was very important to us because we had deadlines to reach. I was like, why don't we use a data pipeline or something that can automate that AI data entry process for us? I went on a little hunt to see if we can find a tool that already does it. We had some tools that did it, but they weren't as accurate as I wanted them to be. Kind of fell apart and we couldn't exactly do it, so we had to rely on the manual process and then use some of the tools that would do the part. And then one of my best friends, she's also an AI developer, she found out there was this hackathon going on, and she was like, do you guys want to participate? So it was me and two of my other friends, and I was like, you know what, let's do it. And then I was like, you know, I came across this problem statement, which is data entry using AI. Why don't we go ahead and try it? So that's how we kind of like decided to build a little mvp. And it actually, you know, came to be pretty well, I guess.
- [04:21]
A
That's amazing. That's amazing. And so now you can, like, take it to the company and say, let's use my product. Tools are messy. That's great. So tell us exactly what is Datrix?
- [04:35]
B
Okay, so as I mentioned earlier, it is a data entry automation tool. So with data entry, we also have like a data analysis part where you can, you know, fetch data for reporting and forecasting as well.
- [04:46]
A
Okay, got it. And I normally ask this question later, but let's go ahead and get into it. How does it work? So if we were to open up the hood, look at its brain, how does it work? And then elaborate to tell us, is it like a patch that goes into our email, is an app on our phones? Just tell us all the inner workings.
- [05:10]
B
Right now it's like a web SaaS platform. So what it does is you can log into Datrix like a normal web application, and then what you do is you can connect it to your Gmail. So we give you like a little code piece. Because Gmail is like, very secure, you cannot directly connect it. So we have like a little code that access your email accounts and we have like A team that would do it for you. So once you connect it to your email, you can also connect it to your data source. So right now we have integration of Airtable, so you can connect it to Airtable. And after that, once it's done, we have an AI agent that goes ahead and, and all of the POs, any order details, any of those basically PDFs or receipts, you get that you need the data entered into your database. The agent would go ahead, study the thing, and it would then go to your database and pick out the schema. So like column names and stuff. And based on that schema, it goes ahead and matches the information and then inputs the data for you. So rather than opening those documents yourself and entering data, AI does it for you. And then you obviously would have some documents that are, you know, maybe previous documents that you want to enter, like PDFs or Word documents or CSV files. We also have like a chatbot interface where you can add those documents and just drag and drop situation. And the AI does the pretty much the same thing. So the agent goes ahead, gets the schema and then gets the information from the documents and then it kind of like inputs all of the data and the thing. So that's like the data entry automation part. And then for the analysis part we have like a, we wanted it to be a bit fun. I think it's more of like a girl thing. So have you noticed like we have these blackboards, like girls usually have them, where they put these sticky notes in to remember like important information and key information. So for analysis Dashboard, what we did was we designed the UI to be like a sticky note information so you can enter like a question chatbot, so you can ask like maybe what is the sales forecast for ABC product? And then it would give you that forecast in a form of like a, either a bar chart or like a pie chart situation. And you get like a sticky note. And then you can customize that, you can make it big, small, you can have multiple sticky notes. So for example, if I have like a presentation where I need like certain numbers, I can just add the prompt. I would have a bunch of sticky notes. I can rearrange them, drag and drop them, delete them, add more. So it's like that's how we kind of get like the data analysis because we don't want those lengthy reports, right? We want to the point information. So that's how the analysis chatbots get.
- [07:28]
A
Into oh my God, that's amazing. And who's your typical customer, right?
- [07:35]
B
So we're targeting majorly right now the retail market or people that have like orders to people that sell and purchase stuff, basically. Right. So anybody who has like customers that basically involve selling or purchasing of goods, whether it be like textile food or like electronics. So we're kind of targeting them now.
- [07:55]
A
What if I'm like a sole proprietor that sells stuff on TikTok, would I be your customer? Are you looking more for the larger retail customers?
- [08:03]
B
So this is like, because Day Trick is like in a very initial stage, we're targeting smaller businesses and then we're looking to move up to enterprise level. If you have like a database and if you have emails that like your purchase orders come comes in. So just pop in. Yeah, you can use it.
- [08:17]
A
And a lot of businesses have that they, you can go and buy but then they get an email and then they have to then go and like, you know, like parse it or copy and paste. So that would, oh my goodness. That will save a lot of people time. That's amazing. Okay, so tell us. I know this is in the MVP stage, but tell us a time where you know, a potential customer, a user, a tester experienced Datrix and it like changed everything for them.
- [08:47]
B
Okay, so what I did was basically my. We started off like in a very smaller testing phase. So my dad owns like a construction chemical company here. So he had like a few orders coming in. So I was like, do you mind if I test it initially? So he's a very huge critique. I would like to say as politely. You know, he's very nitpicky. He's like, oh, this is not right. This is not right. So when I kind of like poised his purchase order and it went to the database as accurately, he was pretty impressed. And then I went ahead to this local bakery that we have here and I just went to them and I was like, hi, do you want to maybe go ahead and try it? It's just like a one try thing. We would be happy to do it for free. You know, we're just looking for potential customers and trying. So they went ahead and you know, tried our product for like just two days, but they were pretty happy with it. We had some feedback as well, like for analysis. They wanted a few more features. Maybe if we have this, that would be really nice. Do you think we can integrate it to more databases and stuff? And we were really happy to, you know, go ahead and help them as well. So it was one of them. Other than that, we also did like little TikTok situation where we promoted it on TikTok. And we did it on Twitter as well. Like, we did, like, a little promotion thingy for a hackathon, and then after, like, testing, we kind of, like, stopped signing people up. But we had, like, over 50 people sign up for a wish list, so that was really exciting.
- [10:02]
A
Oh, that's good. That is excellent. No, it's having kind of wish list or a wait list is amazing. Now, can it tap you. You mentioned Gmail. Can it all. Does it also work with Outlook, or is Gmail just pretty much the main, you know, email tool right now?
- [10:16]
B
Yeah, so right now it only works with Gmail because we're kind of still in, like, an MVP phase. So we want to perfect, like, the features first, and then we would go ahead and integrate with other services as well.
- [10:27]
A
Yeah, no, very smart. And so as I'm having the discussion, I like to point things out. So here, as you know, those of you who are listening, I think that's a really key thing, particularly when you're trying to develop and launch your product. And that's around, you know, trying to perfect it with the one or the most widely used tool before you then, you know, implement it throughout so you can kind of work through the kink. So I love that. Now you gave us an example of working with your father's company, and they're really, really, really enjoying it. So tell me, as you've gone through the testing phase, as you've gone through the development phase, talk to us about a time where you were wowed by what it does and you got goosebumps. And if you have victory dance or victory, like, whatever it is, like, how do you celebrate? Take us to that moment where you were like, oh, my God, like, my best friend and I founded this, and it really is making a difference for folks.
- [11:21]
B
Okay, so for me, the wow moment was how simple this idea is and why did it not exist? So the first time we tried doing, like, the integration with the email, that was kind of like the hard part, because email is very tricky to integrate to. And unless you're, like, a huge thing and you have, like, a huge team of engineers, we were just three kids trying to, you know, play around and try something. So once I finally. So I like, coded the script for the email integration. So once I finally did it and we tried a few, like, you know, dummy POS and what do you say, like, purchase orders, and it added to the database very correctly and accurately, that's when I realized it can be huge, because when I was bringing it up to the client, I Was like, why don't we have something that can do it? He was like, oh my God, that would save us a lot of time. And he mentioned it that they have, they spent around literally millions per annum on data entry and data automation. That kind of the simplicity of this idea kind of gave me goosebumps and the idea that why nobody actually thought of it by now. And I should, you know, bring it up. I can bring, you know, when I, when I say like I want to bring change to the world, I don't mean like I need flying cars and like transparent houses and all of those things. I want to make life simpler generally. Do you know what I mean? Like how chair was invented, people used to sit on the floors and cushion. Then somebody was like, why don't we get a chair? Chair just was that simple thing. I want to have that simplicity that kind of like helps people. So that's when I was really wowed by this. It's like simple, but it would be so useful and it would be so powerful to use if we make it to the industry.
- [12:51]
A
Yeah. And particularly like you said, when you're working on these projects and they're deadlines and you're like, we literally have a trillion different data pieces that have to be entered. Why not use something like daydreams? Got it. Now I'm always thinking about like ethics because this is a lot of data. How have you and your co founder thought about like ethical guardrails and keeping ethics at the forefront?
- [13:17]
B
Yeah. So what we're trying to do is we're trying to make sure that we do not use like reinforcement learning, which is a process where the LLMs or any type of like model that is being used learns from the data because we do not want the client's data to go into these public LLMs. Right. So what we do is we have a pipeline where it just takes that data and it just inputs it. So it's just like kind of like implementing and working on that data. Other than that, because we want to make it to international market. We're currently studying like compliance issues and how we can get certain certificates and any of the cyber security things that we need to integrate with it. So we're kind of like still studying and making sure we kind of like cover all those parts. But right now the first thing we're trying to do is we're trying to make sure not to use any of the reinforcement learning models.
- [14:00]
A
Okay, I love that, I love that. Now does it train on its own data like just this customer, or is it just purely input and doesn't train, you know? Okay, got it.
- [14:11]
B
Does not train. It does not need to train because its job is to pretty much like find like, so basically, like, if I was to explain it in like a very layman tech thing, what it does, it. It goes to your database, right? It gets the column name, right? So let's say the column name is customer name. So it would go to your P.O. it would search for something similar. So it converts it into a vector, which is like a mathematical number. It goes to that PDF, would look for similar vector number situation. So when it finds like the similar number, it goes in, it's like, okay, customer name. Let's say the customer name is Nermeen. It would go ahead and fetch that name and just input it. It does not need to really learn that data.
- [14:48]
A
Okay, got it. Love that. Okay, awesome. Where Nermeen, it was so amazing to have you as a guest on the show. Thank you for being here, everyone. Remember, until next time, lead with AI. Thanks for tuning in to Lead with AI. I'll see you next time as we continue exploring the cutting edge innovations shaping AI across the public and private sectors. Until then, keep leading with AI.