Vasilije Markovic, Cognee, on Reliable AI — Interview with Alexy

Hi, I'm Vasili. I'm the founder of Hi, I'm Vasili. I'm the founder of Cogni. Cognney is a memory tool for AI Cogni. Cognney is a memory tool for AI apps and agents. And uh yeah, that's apps and agents. And uh yeah, that's all. I think the professional highlight all. I think the professional highlight uh involving AI was the ability to uh involving AI was the ability to participate in a lot of events here in participate in a lot of events here in San Francisco where I took the stage and San Francisco where I took the stage and was able to discuss what AI memory is, was able to discuss what AI memory is, what context engineering is and how can what context engineering is and how can we actually make AI more reliable and we actually make AI more reliable and give it the accuracy and access to the give it the accuracy and access to the company's data so we can effectively company's data so we can effectively utilize it to the full extent and need utilize it to the full extent and need it has. Great question. Um reliable AI it has. Great question. Um reliable AI for me is a system that allows us to for me is a system that allows us to interact with the data points and uh interact with the data points and uh other uh let's say elements uh connected other uh let's say elements uh connected to uh operational efficiency. I'll need to uh operational efficiency. I'll need to do this one. So uh reliable AI for me to do this one. So uh reliable AI for me is a system that has the ability to is a system that has the ability to interact with the data from the company interact with the data from the company side, the business side, but also from side, the business side, but also from the data with the data from the agentic the data with the data from the agentic side and merge all of those into side and merge all of those into representation that can learn uh representation that can learn uh activate itself, modulate its own activate itself, modulate its own behavior and uh be um predictable in a behavior and uh be um predictable in a way that we can know how, what and where way that we can know how, what and where happened and why certain answers came happened and why certain answers came from these AI systems. I think right now from these AI systems. I think right now it's a black box and we definitely want it's a black box and we definitely want to change that. I think community uh to change that. I think community uh needs to start building uh because often needs to start building uh because often enough uh the community tends to stay on enough uh the community tends to stay on very high level discussions on what very high level discussions on what reliable AI is about. I think do going reliable AI is about. I think do going to the to the field working on the to the to the field working on the problems trying to solve it for everyday problems trying to solve it for everyday user can actually get you much further user can actually get you much further than than just theory. I don't know but than than just theory. I don't know but with that I would assume that we would with that I would assume that we would have some type of an underlying set of have some type of an underlying set of inference tools like base 10 or others inference tools like base 10 or others that host the models that you would have that host the models that you would have tools on top or systems that allow us to tools on top or systems that allow us to manage the traditional data because manage the traditional data because there is still going to be a need for there is still going to be a need for deterministic data systems. there would deterministic data systems. there would be uh a need for uh let's say be uh a need for uh let's say application layer uh automation where we application layer uh automation where we can launch and delete apps on the fly can launch and delete apps on the fly for everyday use cases and in between for everyday use cases and in between that there is going to be a need for that there is going to be a need for memory and on top of that probably memory and on top of that probably self-improving memory that's what we are