Introduction
On Episode 6 of Bits and Atoms, Elias Can speaks with Dimitris Neocleous, Co-Founder and CEO of iMe, about the growing challenges of building reliable AI systems and the infrastructure needed to make them work at scale. Dimitris explains how iMe positions itself as the “orchestrator of orchestrators,” creating a deterministic control layer that sits on top of probabilistic AI models to address issues like hallucinations, accuracy, and auditability in enterprise environments. The conversation also explores Dimitris’s unconventional path from legal studies to startup founder, why many application-layer startups may struggle as large model providers continue to release competing features, and how to build resilient startups in an increasingly competitive market.
Interview
Elias Can: Hello and welcome to Bits and Atoms, the Grishin Robotics founder series. I'm your host, Elias Can, and today I've got Dimitris Neocleous, Co-Founder and CEO of iMe, joining me. To get started, Dimitris, would you mind sharing a quick intro about you and iMe?
Dimitris Neocleous: Absolutely, thank you for having me. I’m the Co-Founder and CEO of iMe. A few words about iMe; it’s not a model, it’s not an assistant, it’s not an agent builder. It’s the orchestrator of orchestrators, meaning that we bring the deterministic logic of any company as a control plane on top of the probabilistic model and it is industry agnostic. What does that mean in general? It means that right now many internal teams and companies are deploying AI tools but they are hitting the wall when it comes to scalability, they have issues of hallucinations, accuracy, and auditability…this is where iMe comes in. It is a modular solution in order for you to have a safe, trusted, and accurate AI solution in your business.
Elias Can: Fantastic. We’ll definitely get more into product and market, so I’m excited to chat more on that.
Elias Can: Taking a step back and looking at you a bit more personally first, what originally motivated you to become a founder? What’s your origin story?
Dimitris Neocleous: My background is in legal, but I never practiced it. I was doing my legal practice courses in London back in 2019, but I was always a curious person. For me, when I was seeing structure and system I was trying really hard to fit in. So I think it’s my nature to be a founder because a founder challenges the status quo by trying to bring innovative and creative ideas into the market. iMe is my second rodeo, I had a previous startup where we did a fair enough exit. I’m a generator of ideas that I like to design and implement.
Elias Can: Challenging the status quo and finding a problem and trying to build a new solution for it definitely parallels well with the overall sentiment and drive of being a founder.
Elias Can: You said you had a previous company that you exited and now you're working on iMe. How did you land on building this operating system/orchestration layer for AI? What interested you to work on this specifically?
Dimitris Neocleous: I'll tell you how we identified the problem. Taking a step back, if you don’t pivot in your startup, you are not a startup. You are, I don’t know, a public service. In July we did the second pivot in iMe after we collected a lot of feedback and we realized that the problem right now is the fragmentation of data and tools, so we started building the orchestrators. Then we saw OpenAI is doing the orchestrators with Agent Builder, Gemini Enterprise was also doing something similar. It was interesting because it validated what we were building. By the end of September we had our first MVP and I saw a very interesting behavior from AI. We were building a solution for a company, an internal legal assistant, and we were asking some questions to the assistant about finding a contract and suddenly the AI was creating a contract that didn’t exist on the database of the company yet it was showing it like it was created by the company. After that, I had a decision to make…do I be fully honest and communicate this with the client or hide it under the hood and tag it as a misaligned behavior of the AI? I decided to be very transparent, and they appreciated my honesty.
Internally, I said to my team we need to tackle this challenge right now. Not on the side of fine-tuning and guardrails, I think we need to do something much more. Why I said not on the fine-tuning and guardrails is because we have been reading different research papers from MIT, Anthropic, etc. and once the model is getting upgraded the AI misalignment becomes bigger. What do we mean by AI misalignment? We mean the behavior of creating something that doesn’t exist just to please the user or trying to persuade you that it is the right answer. So all of this we are tackling with our modular solution. We decided to focus on this part, rather than creating AI assistants, because it is actually a big market and also gives us more opportunity to work with projects that previously would have been competitors to us. Now we’re becoming their suppliers and we’re helping them go to market faster with more security and efficiency.
Elias Can: Very interesting, you’re building the pick-and-shovel tooling for folks that are building on the AI assistant side.
Dimitris Neocleous: If you think of an assistant as a train, we are building the tracks and we are making sure the train doesn’t derail. Many startups focus on guardrails or fine-tuning or prompting, we took a different approach and said the problem is not the model, the problem is the architecture of it. So we brought the deterministic approach before the model.
Elias Can: So when you’re going to potential clients, is this probabilistic piece the key differentiator or selling point to them to work with you? What is that sales conversation like when you’re approaching a potential customer?
Dimitris Neocleous: What we are doing in our pitching is saying we know you’re building an assistant, we know you’re facing these challenges like hallucination or sycophancy, or even the creation of data that doesn’t exist. So we have them do a pilot, and if they are happy with that then become a client. They have nothing to lose because it’s a pilot and during that time we’re proving the value of iMe.
Another thing we are doing is we are creating benchmarks that we plan to open-source. This way any startup’s internal team can test themselves based on our benchmarks. We’re using different third-party auditors to help us and identify those benchmarks.
Elias Can: Sharing those benchmarks is probably a good narrative or public relations piece for you as well.
Elias Can: Maybe a broader question here, and one that's a little bit forward-looking because AI is top-of-mind for everyone right now. How do you see AI evolving? What’s your view on the current landscape of early-stage AI startups? What does the market look like in a few years from now?
Dimitris Neocleous: I'll tell you my view and it is very cynical, but at the same time I’m trying to push the limits. I think my generation and the generation after me is a bit comfortable in the sense that they have stopped being creative. A lot reminds me of the dot-com bubble. Right now, there are a lot of startups that are building on the application layer. When you build on the application layer, every time OpenAI or Anthropic releases something they’re killing a bunch of startups. For example, Anthropic recently announced Claude Plugins and it helps with productivity across a number of domains. Just with a plugin, Anthropic is now competing with any startup across those domains and just showed how easy it is to build it. Investors will begin to wonder, “hmm, maybe we have put our chips in the wrong startup or in the wrong project.” Anthropic also announced Claude Code Security, and we immediately see a downtrend in public companies like CrowdStrike and cybersecurity companies. This shows the old SaaS model is under threat. This also shows the focus on one problem can be a good idea, but it cannot be a shallow problem. For example, if you’re focusing on productivity for legal, and it’s on the application layer, I can guarantee you that 99% of those companies are going to die from Anthropic, Gemini, or OpenAI releases. Those companies, they have or have raised so much money and are trying to make it as sticky as possible to their platform. Otherwise they are going to go bust, and there is not enough liquidity. So what are they doing? They are killing the existing SaaS and they are killing a broad number of startups. This is where I’m seeing the issue and my suggestion to every startup is to think of deeper, not shallower, problems. Now is not the moment of shallow problems because the barrier of entry for any tool is much less. Any non-coder can create something, before being a coder was a valuable niche. You either need to go to the infrastructure layer or you change a business workflow so that your solution will make sense and be sticky enough.
Elias Can: It’s both a spicy take and some practical advice for folks out there. It’s going to be interesting to see how this all plays out. To your point, every time OpenAI or Anthropic releases a new update they’re killing swarms of startups…it’s challenging. Apple kind of had that same ability in the past where if they built something new natively for iPhones it could kill single applications in the App Store. So maybe we are seeing not exactly a repeat of history, but more of a rhyme.
Dimitris Neocleous: And it's a faster time because the friction for non-developers to put something in the market is much less thanks to vibe coding tools. It’s the Wild West per se, and the speed of things is literally crazy.
Elias Can: There’s never been a time where things are easier and faster to build and ship. Even just keeping up with the high-level AI news on a weekly basis is a full-time job in and of itself.
Elias Can: One last question here, what's been your best strategy or strategies for finding potential investors so far? What does fundraising strategy look like for you?
Dimitris Neocleous: Every quarter the strategy changes. Since we’re living in very uncertain times, geopolitical situations etc., but also with regards to what I said about OpenAI, Anthropic, Gemini, etc. Europe has always been a bit more conservative in terms of investing in startups, but now the US is also to a degree because of how much those big AI companies have raised already. We are seeing investors becoming a bit more conservative in general. You can’t really raise an early round on a vision and be pre-MVP. You need a good network and may need to go to an accelerator with good credentials as well. I think more founders will need to bootstrap or just raise enough money that will keep you afloat for the next 9 to 10 months. It sounds like gambling, but if you really believe in your idea and your team it’s not gambling…it is betting on you. Then you have something solid to show either with good traction or good benchmarks. Bootstrap as much as possible and find investors that believe in you. A small check can be more important than a big check with constraints and pressure and unfortunately the startup world is becoming F1 where it’s a very expensive sport.
Elias Can: I think you hit a few good threads here. One, investors are looking to see more and more traction early on, they need to see proof points because there’s much less friction for building a product nowadays. So the bar for traction is higher. Two, we are seeing a lot of bifurcation in the market as well where some folks are raising really big rounds early on and it’s unlike anything we’ve seen before where the first round or second round of a company is nine figures plus. People that are coming from OpenAI or Anthropic who have name brands are raising these massive pre-seed and Seed rounds. Then on the other side, you have a lot more founders who are bootstrapping or raising smaller amounts first to find great traction narratives and use that to raise an amount of funding that works for them. This also means they don’t have to become a unicorn overnight and put that immediate pressure on their business.
Dimitris Neocleous: Right, don’t raise money just for the sake of raising money…raise enough for the next 10 to 12 months. I know this sounds scary for founders, I’ve been there, but the survival model kicks in. There’s a rush of intuitive creativity when you are in survival mode. You see the whole team working hard and it is beautiful.
Elias Can: Thank you, Dimitris, for joining me on Bits and Atoms. It was great to have you today and get your perspective here. I'll let you get back to building.
Dimitris Neocleous: It was a very interesting conversation, one of the few that was unfiltered and uncensored. Also thank you guys for the GRx Accelerator in London, it really changed my life and I’m very grateful for it.
Elias Can: We loved having you and glad that it was able to have a good impact on you! We're very grateful to hear that.
Dimitris Neocleous: Thank you very much!
iMe is an enterprise AI operating system that solves the critical governance and control gaps preventing organizations from moving AI from pilot to production. Built on a hybrid deterministic-probabilistic architecture, iMe provides the governed rails that enterprise AI needs to operate reliably at scale.
Grishin Robotics is a Silicon Valley-based VC firm focused on investing in early-stage companies in broader consumer categories. Grishin Robotics has invested in category-defining companies such as Ring (acquired by Amazon), Spin (acquired by Ford), Zipline, Starship, and many others.
