The Proof of Concept Worked. Then the World Changed.

May 2021. Two interns. One question.

Not how do we build this. That came later. The first question was simpler and more uncomfortable: can this be built at all?

I had no technical background. No co-founder. No funding. What I had was a clear picture of the problem: domain experts locked out of their own data, permanently dependent on someone else's calendar. And a working theory that a single no-code platform could solve it. Five modules. One interface. Point and click. No SQL required.

The interns were CS students. Smart, willing, patient with a founder who couldn't always tell them exactly what he needed, only what he was trying to achieve. We had Stack Overflow and tech documentation as our only guides. No AI tools. No shortcuts.

The mission wasn't to build a product. It was to answer a question: could five data modules (connect, replicate, transform, visualise, predict) actually live together in a single no-code interface? And if they could, why hadn't Google or Microsoft done it already?

We spent two and a half years finding out.

What Those Two and a Half Years Actually Looked Like

Building software without a technical co-founder is a particular kind of slow. You learn what you don't know by failing at things you didn't know were hard. The sequence of problems is never what you expected. Progress looks nothing like a roadmap.

By late 2022 we had something that connected data sources. By mid-2023 we had pipelines that actually ran. The interface was rough. The security was naive. The UI looked like a proof of concept, because that's exactly what it was.

What made those two and a half years survivable was the feedback loops. Not encouragement. Actual pressure from people who had skin in the game of telling the truth.

Throughout this entire period, every few weeks, I would demo to a mentor and former colleague from an earlier chapter of my career. Someone who had gone on to senior engineering leadership at a well-funded AI fintech, built by founders he had worked with before. He was not encouraging. He was useful. Real engineering feedback. Specific gaps. Hard questions. Pushback on what wasn't working. He had been inside two serious AI product builds and he gave us the kind of pressure that made the work sharper. Every session left me with a clearer picture of what was missing and a less comfortable picture of how far we still had to go.

Beyond that, we demoed repeatedly to a handful of startup founders. People not yet in a position to pay, but living the exact problems the platform was built to solve. One of them was running a real, operations-heavy business with genuine data problems and no dedicated data team. He represented exactly who we were building for. His questions were never about the technology. They were always about whether the answer he needed would actually be there when he needed it. That kept us honest about what mattered.

Every feature priority, every architectural decision, every painful rebuild: shaped by these conversations. Not invented from my imagination alone.

November 2023. MongoDB as the source, BigQuery as the destination. All five modules. Naive, unpolished, not ready for any real user. But working.

The answer to the question was yes.

The Thing Nobody Talks About When ChatGPT Changes Everything

The week we confirmed it worked, the conversation everyone was having was about AI replacing everything we had just built.

ChatGPT had arrived a year earlier and by late 2023 it was everywhere. The headlines were daily: AI is killing data science. AI is replacing analysts. AI is making the entire analytics stack obsolete. Why would anyone need a platform to query and transform data when an AI could just answer any question you typed? Gartner predicted in October 2023 that more than 80% of enterprises would have used Generative AI APIs or deployed GenAI-enabled applications by 2026 (Gartner, Top Predictions for IT Organizations and Users in 2024 and Beyond, October 2023).

I spent three months genuinely uncertain whether we should continue.

Not anxious. Not briefly shaken. Genuinely uncertain. Demotivated in a way I hadn't been since before the idea had a name. Out of money. Looking at two and a half years of building and wondering whether the thing I had set out to solve had already been solved by something I had no part in.

What nobody tells you about this moment, and every founder building something slow and foundational will eventually face a version of it, is that the doubt isn't irrational. The headlines weren't wrong. AI was genuinely changing what was possible. The question of whether our specific approach was still necessary was a real question, not a crisis of confidence to be powered through with a motivational framework.

I sat with it. I tested the argument. I asked whether an LLM bolted on top of fragmented, unvalidated, ungoverned data actually solved the problem I had set out to solve.

It didn't. It produced confident-sounding answers from unreliable foundations. It hallucinated in ways that were harder to catch precisely because they sounded authoritative. The domain expert asking the question had no way to verify whether the answer was correct. Which was exactly the dependency I had spent two and a half years trying to eliminate.

Why the Foundation Had to Come Before the AI

The three months of doubt forced a sharper answer to a question I had been circling since the beginning: what actually has to exist before AI can be trusted to answer a business question?

Not the AI. The foundation underneath it.

Clean data. Validated data. Governed data. A semantic layer that gives an LLM the context it needs to query correctly. Encryption architecture that ensures no raw data ever reaches a model. An audit trail so a CFO can defend the number in a board meeting. That is what Integrate and Visualize are built on top of.

Everyone else in 2024 was bolting AI onto existing fragmented stacks. A chat interface on top of unvalidated data and calling it intelligent analytics. Edilitics couldn't do that. There was no existing stack to bolt onto. So between 2021 and early 2024, we had built, without fully knowing it at the time, the only thing that makes AI analytics actually work.

When I collected money from friends and family to keep going, the thing I was funding wasn't just continued development. It was the decision that the foundation mattered more than the speed of getting to market. That the infrastructure, the security posture, the semantic layer, the governance architecture: all of it had to be right before any AI layer could be trusted.

That decision is why AskEdi works the way it works today. Not because we were smart enough to plan it. Because the crisis of late 2023 forced us to be honest about what we were actually building.

Why Every AskEdi Answer Shows Its Query

In May 2026, the AI analytics market is full of products that went fast and skipped the foundation. Chat interfaces on top of unvalidated data. Confident answers with no audit trail. Governance bolted on as an afterthought because enterprise customers demanded it.

Edilitics didn't skip it. Not because we were visionary. Because we had no choice. A non-technical founder building without a co-founder, learning in public, failing in private, couldn't afford to build anything that didn't actually work. The proof of concept had to prove something real or there was nothing to carry forward.

The three months of doubt in late 2023 are the reason every answer AskEdi returns shows the query that produced it. The reason your raw data never reaches a model. The reason a domain expert can ask a question and trust the answer without a data team standing between them and their own numbers.

What the market is now scrambling to retrofit (the governance layer, the encryption architecture, the semantic foundation, the audit trail), we built because we had no other option. Not as a differentiator. As the only path forward that made any sense given what we were trying to solve.

It nearly ended everything. Instead it clarified everything. And that clarity is what the next three years of Edilitics is built on.

Sources

Gartner. Gartner Says More Than 80% of Enterprises Will Have Used Generative AI APIs or Deployed Generative AI-Enabled Applications by 2026. October 2023. Retrieved May 2026. gartner.com