You may have seen the announcement. This is the part behind it.

Tirth and I have known each other for twenty-two years. That number felt strange to type. But if you count from college, that is where it lands.

It started the way most tech friendships do — long arguments about things that probably did not matter as much as we thought they did. How classes and objects work. How pointers manage memory. C# versus Java. The kind of debates that feel urgent at twenty and hilarious at forty, but that quietly shape how you think about building things forever.

Then life moved the way life does. He went to TCS, then Intel. I went through Patni, Capgemini, HP. When we both ended up in Bangalore for a stretch, the arguments just upgraded. Now it was cloud and IoT. How operating systems actually work under the hood. How databases find entries faster than a file search. And, memorably, a long conversation about how to not crack MNC interviews — a topic we had more practical experience with than either of us liked to admit.

Then the paths diverged more seriously. He went to Cornell for a master's in Machine Learning, on scholarship. I moved into consulting and eventually founded Fuzzy Cloud. Different cities, different time zones at points, but the underlying rhythm stayed the same: whenever we had time, we found ourselves back in the same conversation. Machine learning, cloud infrastructure, whatever had just come out that week worth arguing about.

Where Legal Desk AI Enters

When I started building Legal Desk AI, I carried a specific conviction into it: lawyers deserve privacy-first AI, not privacy-as-afterthought AI.

I am aware that many lawyers do not yet fully understand the privacy implications of how they use AI tools. But that is not a reason to take shortcuts. It is a reason to be more careful, not less. Deleting files after seven days to manage storage costs and calling it a privacy feature is not a privacy philosophy. It is a cost decision wearing a privacy costume — and in Indian legal practice, where a matter filed today may see its next hearing eight months from now, it actively breaks the workflow lawyers depend on.

The question I kept returning to was not what can we get away with but how do we build something that genuinely performs without compromising what matters. That is a harder problem. It requires real machine learning thinking, not just calling an API.

Those long conversations with Tirth — about RAG architectures, about grounding, about getting that last-mile performance out of a model without feeding it everything indiscriminately — those conversations are directly what became SMRITI. Your personal legal knowledge base. Context that belongs to you, searches in a way that understands how lawyers actually think, and does not disappear between sessions.

Tirth's instincts run toward ML performance and privacy in ways that complement where my strengths are — distributed systems, serverless architecture, cloud infrastructure. We have always been better at thinking together than separately. Once SMRITI shipped and I started working on what comes next — something larger, more details when it is ready — it became obvious that having him officially involved was the right move.

I reached out. Once the formalities were sorted, he was on board.

I could not be happier about it. Twenty-two years of tech arguments, and now this. Some friendships just find their shape eventually.

Welcome officially, Tirth.