Introduction

So Anno as a company, we started out as a data labeling tool, but quickly we became a company whose mission it was to help make AI more accessible. So what does this mean? As many of you guys know, in the past five years, particularly, more than that, there's been all these new research techniques coming out.

BERT, GPT, GPT-2, GPT-3, Palm, Gemini, Sora, Claude, all these new models. You kind of see things like Opera coming out. And there's all these incredible innovations.

And obviously, most people here would agree that AI has the potential to be amazing, to really make a positive impact in our lives. Today, the biggest challenge isn't that the technology is behind. The technology can already do a lot of things. But there's a huge gap between where the technology is today and the everyday tasks that people care about.

Even today, you have less than 15% of adults in the USA that use chat GPT. So think about if we have a population of 335 million people in the USA, probably around 250 million of them aren't even using this revolutionary technology. And even those who kind of are using it just feel like they're falling behind.

Our Mission and Goals

So the goal of this AI Day Summit is a bit different than a meetup that we normally do. Summit's kind of about like action, taking action to make a change that we want to see in the world. And so what we want to do today, and our kind of goal of hosting all these meetups and talks and all the work we've done, to try to help bridge the gap between the amazing potential that AI can provide and the everyday, today-to-day tasks that people care about. So that's really why we're doing this.

And we're obviously going to be launching a few things, but we want this to kind of be a community effort. So we want to do as much of a job as possible to kind of share why we're actually building what we're doing, why we think that is aligned with this kind of mission of helping make AI accessible. In the end, we'll kind of have a call to action. And hopefully, if anything doesn't make sense, just come find me after, and we'll kind of clarify anything.

Current Challenges in AI

So when it comes to the AI space right now, we're going to say that there's five problems that our kind of company right now is trying to solve, which is a lot of problems. And it goes from simple to, in my opinion, really complex. The problems we're going to try to solve are related to the data gap, the AI war, robust AI agents, evaluation of AI models, and the future of autonomous agents. So let's start with problem one, the data gap. So the problem with the data gap is this.

Obviously, there's been this incredible transformational technology that has kind of come out related to generative I and LLMs. So you have these LLMs like GPT-2 and GPT-3 and now GPT-4 and O1, and you kind of need to train these models. And to train these models, you kind of need to use a lot of data. And so people will kind of, to train like Bert, for instance, you take all of the data within Wikipedia and all the data within Google Books, and you use that to kind of embed the knowledge into the model. And right now, we're running out of all the publicly available data. What does that basically mean?

If data is kind of the building block to building these models, there's kind of like a limit to how good these models can get with the data we have. This actually isn't too big of a problem because the model's already good, but if we want to kind of get these models to like next level success to do a lot of these like enterprise use cases that businesses care about, it's really difficult for them to do with kind of like today's current models. But these enterprises actually oftentimes are sitting on tons of data. So the amount of data that's actually used to train a model like GPT is very, very small and minuscule to the amount of data that maybe a company like JPMorgan Chase or the DoD has.

So in order to actually help these companies build great models that can solve all their analytics and enterprise-related use cases, You need a way to leverage their data to train models that can do better than the initial models. That's how you can kind of build great AI systems.

Our Solutions

So our solution, and one of the first things we're going to be launching today, is OpenAnode. So OpenAnode is essentially the culmination, I think, of two and a half years of really hard work. Essentially, it's a free and decentralized way to train your own models on your own data in an end-to-end MLOps-like way.

You can label data, train models, make predictions, evaluate the results, and integrate the best model into your data set. We have this data labeling interface where you can define the ontology you need and convert the really raw unstructured data to like an LLM ready format.

From there, you can actually train the model with a variety of different techniques. And you can kind of take the models you've trained and actually use them to do inference, where you can kind of put the models into your own chat bot, similar to the chat GPT, but it's fine tuned and accurate. And you can evaluate it with a variety of metrics, as well as integrate it to your NTEN workflows with our SDK. We have actually provided most of this kind of available for free.

So if you actually go to our website right now, and you hit this Learn More button, you can kind of get all the information you need for Anote and OpenAnote right here. We kind of share how this kind of works. We have a documentation guide that goes through all the details. We have a product that we're actually letting people try and use completely for free. You just go to this website and sign in, and you get to something like this, and it all mainly works.

The way this kind of works in theory is you can kind of create these projects, define your training and testing data sets, Essentially, define the evaluation metrics you'd want to use to actually train your models. Then from there, you can just basically try all your initial zero-shot models like GPT, CLAW, LAMA, and MISTRAL. You can do fine-tuning with a variety of different techniques.

You can label data to improve the model's performance by adding ontology you'd want. And then you can kind of iterate on the models, get the best model for your data, and get a report of all the models and how they perform and integrate it via our SDK. So that's OpenAnode.

I'd say, in addition to OpenAnode, we've actually put most of the materials that we've kind of built related to our products, such as how our code base works, Different materials related to our pitch deck, marketing materials, case studies, research papers that we've kind of worked on. Readily accessible for anyone who wants to kind of learn about AI and how this works. Just out there and you can just kind of go here and like see all the work we've done for the past few years. So yeah, that's OpenAnode.

The AI War and Its Implications

The second thing was kind of a culmination of what we basically realized when kind of going to market with Anode. So it's clear, at least from our perspective, that right now, if you kind of read the news about DeepSeq and all, there's this kind of battle or war going on between America and China. And it's different than other types of wars where there's a cyber security war or, you know, the race to space. AI is like this insanely powerful technology.

So the winner of like the AI war will essentially have a huge say in what the future looks like. So in America you have some of these leading companies like OpenAI and Cloud. And then in China, you know, there's been this shift towards open source AI with kind of models like Gwen and then most recently like DeepSeq that's taken the world by storm.

It's really important that America stays ahead in the AI race. And right now we're ahead on the algorithms, but that's kind of quickly losing pace as all these open source models are getting better and better. And most people kind of agree that the rate of open source model progress is getting closer to closed source model progress over time.

The big difference and the big... thing that's kind of scary is when it comes to China, they actually have access to a whole bunch of data that America doesn't have access to. So if you also believe that the data used can actually help actually train these really powerful large language models, it's basically inevitable that over time, China is going to get closer and closer to being really strong in AI. And we need to ask America to kind of meet that or match that or exceed that.

So we need to really invest in the data to help power these AI models. And that's why a lot of these folks in the federal space are really looking for different data solutions. And that's why we wanted to share this with you, because this has the tremendous potential to have a really big impact in the future of the AI race.

Internal AI War

Secondly, we're not just in this war that's kind of like this USA versus China AI war. But more importantly, we're starting to see something that's happening today related to this internal AI war happening in the US. So this is something that I feel like most people can relate to, but we're seeing a lot of jobs starting to get automated. Most people kind of would estimate in the next five to 10 years, between 20 to 30 or 40% of the jobs that are like white collar jobs will be like automated with AI.

These could range from things like insurance underwriters, accountants and auditors, medical transcriptionists, even financial analysts, like a highly skilled job. These AI models are actually becoming really good, which is scary because people that know how to use AI are kind of starting to replace those who don't. So the challenge we kind of have for you guys and us is we need to help people stay ahead of the AI curve and learn about this technology. We need to make sure it's really accessible so that people are able to kind of adopt and train and use the technology to stay ahead and not fall behind

which is very important, right? So we need a solution to that problem. We need to kind of help make AI more accessible, not just for people in New York, but all across America, and ideally all across the world. So that's kind of why we host these AI events in New York, and that's why we're launching this new thing called the Armour Institute.

So the Armour Institute, as you guys know, we're going to call this the first chapter of the Armour Institute. But the Armour Institute is a community with AI workforce training, events, and meetups. We want to kind of help make AI more accessible and practical and build a community of consortium of AI researchers and institutes across America. So let me kind of share some more details about what we're thinking to make this a bit more concrete. So if we go to our website and you go to the Armour Institute, you can actually read our general plans, which we've kind of open sourced.

So we have chapters right now in America in coordination of MindStone in New York and Toronto. And where we want to kind of go by the end of 2025 is to have these AI meetups in communities across America. So one place we're actually going to be going to is Boston. Starting on early March, we're going to be hosting meetups at the Microsoft Nerd Center.

there. And the general kind of consensus is for each kind of place you'd want to host a meetup, you essentially need to find a really great venue. You need to kind of know a little bit about the events that are going on there. And you need to also kind of have a list of really incredible speakers that are people that you want to listen to. So we are kind of building this kind of consortium of hosting these AI communities and meetups across America to help actually solve the problem of making AI more accessible.

I'd say this isn't something we're just kind of doing like a half ass. Like we want to have like a plan for like commercialization of a way that we aren't just going to be spending a lot of money, but we're actually going to make an impact and use this to help not just make AI accessible, but also propel America ahead in the AI race. So we're actually partnering with a variety of leading organizations on behalf of smart AI. companies like MITRE, Pfizer, Texas A&M, Airbus, universities like UConn.

And kind of together, once we build this community, we want to work together with people in our community to collaborate on different AI projects. And we're essentially open sourcing a pipeline of a variety of projects that we want to kind of work on together. And that's kind of our plan to kind of bring people into the workforce, learn about AI, and work on these amazing projects as a team. So yeah, that's the Armour Institute.

Productionalizing AI Agents

I personally am really excited about it because it's really solving the core critical problem that our company is meant to solve. The third problem that we're trying to solve is related to productionalizing AI agents. So if you kind of came to our virtual AI Day Summit today, you would have seen some of the most incredible AI agent projects. Things like a data analyst agent, an AI agent to help you kind of improve the job search process, an AI agent to kind of do email outreach, an AI agent to look at financial news and analyze things, AI agents for marketing. We've kind of seen like all these different types of agents.

But right now in the AI space, all these AI agents aren't really in production. Maybe that's like a technological issue, but it seems to me like even though people are building all these really incredible AI agents, there's a big gap between the building of the POCs and the usage of these agents in production. For one, maybe in the next few years there'll be breakthroughs in technology to make this possible. But for two and three, it's hard to even find these different AI agents that people are making.

And there's not really a place to share AI agents that you've built with others. So as we build this community of people across America, and we meet a lot of people who build these AI agents, we want to have a place where they can share the agents they've built. And others can see it and contact them to help get these agents in production in a way that can help people. That's why we're releasing the Agent Registry.

The Agent Registry is an AI agent hub to find and share high-quality domain-specific agents. We, as a team, have built the first six agents in the Agent Registry. One of the agents we built is Upreach, which essentially helps you find leads and reach out to them. Spirithy will talk a little bit more about that. We built an AI-assisted RFP agent to help you apply to RFPs.

We built an AI coding agent, which Aisha showed today, that can help you get your initial code in place to make PRs. We built a financial chatbot agent that can look at information from the 10-Ks and answer questions. We built, obviously, a semi-newsletter agent. But people in our community have built hundreds of these agents.

We even talked today probably to 30 people who built agents. And the main problem is people are building these things, but nobody really knows about them. So we need a place where people can share their agents. And as we build this community, they can benefit as well.

So we're going to have a simple process that we're going to publish on our website on Valentine's Day, February 14th. for people who want to add agents to the registry. It's kind of simple. You just kind of create the agent card. It looks something like this, where you can have an About section, the features, the use cases, the capabilities, probably like a short YouTube video demo of how your agent works, which anyone who kind of comes to our AI Day Summit and presents has that video.

Then you can basically submit an agent card to add the agent to the registry. From there, that will basically go to my inbox, where our team will review the AI agents you share, ensure that they're high quality, the things that we think people would want to see. And then we're going to essentially post it on our website right here, where you see Agent Registry. And you'll be able to see your agent here.

And you can click on this. You can view a demo, which will go to a YouTube video that I won't show of Christie. But yeah, you know, that will be it. And that way, everyone can kind of win as we build this kind of community across America.

Evaluation and Collaboration

And you can kind of share your projects with others. Kind of simple, right? The fourth thing, this is more of like an AI and ML related focus thing that not many people know about until you get into the space and then it becomes like really obvious that it's really critical. It's a problem about the evaluation of AI models.

So something that we've kind of seen as we've built our Anno product is any kind of person we work with, there are some kind of clear evaluation metrics where people can maybe wanna extract entities or summarize text or classify answers to questions. And for those use cases, we have a product that works okay to pretty well at doing those things, right? There's obviously dream improvement, but it's like one of the best out there. And then I'd say 70% to 80% of the use cases that we actually get are things that are really cool projects, but our product can't support. Things like looking at videos and transcribing them, or doing multimodal RAG, or doing things like agentic multi-agent evals, and building agentic AI models.

we kind of get a lot of requests from people who want to do these things. And we have also an amazing community of ML engineers and practitioners. So there needs to be a way that when we get a kind of a project or use case from enterprise that people want to do, we can kind of connect them with the ML engineers in our community to work on these really exciting problems together and to win if they do well. So we're releasing the Model Leaderboard. The Model Leaderboard essentially is a marketplace between enterprises or small to medium-sized businesses who have a data set and want to solve a problem, and AI and ML practitioners who actually build AI and ML models to do really well in those eval sets.

So our team, including Spurthy and a few others, actually built the first six data sets on our model leaderboard. And the way it kind of works is when an enterprise can kind of submit a data set to us, or if we have anyone in our community has a problem they want to solve, they can kind of submit their data set to the model leaderboard. And then we will kind of, check it out, we'll kind of review the data sets and make sure they're good. And if it looks good, we'll add it to our benchmark data sets in our product.

And we'll kind of share a link to kind of like learn more in our Slack channel as well as our like, hopefully newsletter soon. And then people that are AI and ML practitioners in the community can see these data sets, they'll know that kind of they're released. And they can actually submit their models on this data set that they care about to the leaderboard. which we'll review with the enterprise to ensure the models look really well. And then if it looks good, we'll actually add their model to the leaderboard to see how they performed against others.

What's cool about this is if you submit a really great model to the leaderboard, there's a strong likelihood that the end customer that will go to us might want to work with you guys to help convert that POC model to something that they can use in production. So that's the model leaderboard.

The fifth problem, it's a little bit more of a far-reaching thing. And Spurthy will kind of talk a bit more about this. But it's one of the most pressing problems today.

The Future of Autonomous AI Agents

And It's called the future of autonomous AI agents, but this should probably more likely be called the future of really powerful AI. So right now, given this major problem in regards to the accessibility of AI, where you have just a few people that are building all these incredible models, when it comes time where these really smart and incredibly powerful AI systems are in the hands of others, and they get better and better and better, if we only have a few people in the room to kind of have a say in what happens, that would be really bad.

So if you believe that, hey, AI technology is gonna become better and better over the next few years, you would also agree that it would be a good idea to have as many people in the audience in our community have a say in what this technology will look like. So we need a way that is open, free for anyone to contribute But we can kind of all learn together about these kind of incredible technologies and AI agents so that we can build a world where it's not just the people with the biggest brains or the best coding have a say in what these agents look like, but it's also everyday people with big hearts that care about others.

So that's why we're going to be releasing something that's still in the works, but it's autonomous intelligence. Autonomous intelligence is essentially something we started two months ago, so it's still pretty new. But it's an open source research project aimed at building collaborative multi-agent AI systems.

And I'll let Spurthy kind of talk most about this, but the key thing is it's a really powerful AI technology where you can have these teams of agents that work together, similar to how humans do, to build these incredible systems that can do a lot of tasks in a generalizable way. So it's different from the agent registry where those are like specific agents, like the data analyst agent. But in the autonomous intelligence is a generalist AI agent framework that could be really powerful.

And we want a lot of people to kind of be involved in that project.

That's a lot of information.

Conclusion

I guess to close, we've kind of built a company called Anno where our mission is to build accessible AI that empowers anyone everywhere. We want to kind of build a community around that to help bridge the gap between the tremendous potential of AI models and everyday tasks that people care about.

We're really thankful for you guys taking the time to be here today.

We'll be having our formal launch announcement on Valentine's Day, February 14th, but here's like the QR code if you're interested in kind of learning more about each of these initiatives. And of course, here's a link to join our Slack channel if you want to kind of be involved in any of these products.

Thank you.