On Not Nullifying Years of Effort - A Mid-Journey Reflection

I’m writing this as much for myself as for anyone who might stumble upon it. Sometimes you need to have a conversation with yourself, to address the voice that whispers “you haven’t done enough” when you see others building research labs or scaling companies.

I think I’m a pretty unique breed in this ecosystem. Not because I’m special, but because I often find myself overwhelmed by the question: “People are building research, people are building companies – why am I not able to do that?” It’s a question that haunts many of us, I suspect.

But here’s what I’ve been reflecting on lately: we shouldn’t nullify our years of efforts and years of learning. Whatever we build next should be built on top of that foundation, not from scrapping everything and starting over.

The Work That Often Goes Unseen

Let me be clear about something – I’m not just someone who talks about AI on LinkedIn. Over the past six years, I’ve worked with companies, built models, written code, deployed solutions, run business calls, closed deals, set up teams, and trained people. It’s not the typical influencer thing where you just discuss papers.

In Tamil Nadu – a place where very few companies are doing actual machine learning – I led a startup’s AI team development. I trained frontend engineers to learn AI. We worked with IBM as vendors, validated their solutions, and when things weren’t working, we built our own. We prepped data, created a dataset of 30,000 to 40,000 annotated documents, and trained models starting from small VLMs and moving to vision language models. We worked with Qwen 2.5, validated continuously, and explored parameter-efficient fine-tuning.

The result? We established an ML practice in a company that had nothing before. Not just talking about AI, but actually having clients, developed models, evaluation pipelines, and teams that work – all happening in tier-three towns in deep South Tamil Nadu, not even Chennai or Coimbatore. I built the team, set the narrative, established the practice, and now people are doing the work.

This isn’t just one place. It’s been everywhere – Informatica, Captain Fresh, Mu Sigma. But sometimes, strong perceptions make you forget all of this and think, “Oh, we haven’t done anything.”

A Day in the Current Journey

Just yesterday – in a 24-hour timeline at Columbia – I was reading and working on replicating sparse autoencoders to do interpretations of small language models. It’s not a new thing, but it’s something I’m learning and trying to apply to specific use cases. I’ve been reading Neil Nanda’s work, trying to replicate it.

Meanwhile, I’m collaborating with a materials science lab and statistical risk analysis lab here, working with PhD students on conceptualizing new ideas. We’re figuring out what apparatus to select, what models to use, what data to work with. We’re moving toward literature surveys and building frameworks.

My coursework demands that I work on mathematical foundations – the theoretical parts I’m trying to understand. Yesterday, I had a one-and-a-half-hour conversation with a location intelligence company that’s just getting started. They wanted to discuss high-level problem statements, how to approach them systematically, what data sources to use.

And then I’m back to coding, testing models – sometimes scrappy work here and there because I have 24 hours and about 12 of them are fully focused on studies and collaborations. I’m learning, looking for the right distractions and nudges, interesting problems to work on.

The Gap I’m Trying to Fill

Here’s what I’ve realized: not everyone is meant for pure research, and not everyone is meant for building enterprise solutions. I’ve been able to play around in different spaces for six years, but the stickiness hasn’t been there. I don’t have a solid research profile or a solid open-source profile. That doesn’t mean I haven’t done anything.

I’m not denying that I need a better research and open-source profile – that’s exactly what I’m trying to build now. When I’m able to do that, I believe I’ll succeed in this aspect too.

I see myself standing at the intersection of applied work and research – interdisciplinary, bridging the gap. I’m not doing pure foundational research, and I’m not just an applied professional. I’m something in between, and I need to prove that more visibly.

I’m also not doing the influencer thing on LinkedIn – no rage bait or clickbait. I’ve started sharing my thoughts on YouTube, but I’m not fully turned into a YouTuber either. It’s interesting how people who are consistent at very boring or silly things often set the narrative better than those who are truly involved in addressing the ecosystem’s real needs.

What This Journey Is Really About

The thing is, people trust me. They come to me with ideas, book time on my calendar to discuss agent AI for meeting facilitation, or how to approach specific problem statements. Sometimes I worry I’m seen as someone who just talks – “Oh, let’s use this algorithm, this methodology” – but honestly, that’s common sense if you talk to most machine learning professionals.

What I’ve noticed is that people sometimes can’t believe how I can talk different languages – not literally, but at a sensible level (not expert level, but sensible) about how a problem could potentially be an analytical or data science problem versus an AI problem, or how we could approach applied research or theoretical aspects systematically. Just yesterday, a conversation at a coffee shop with my junior helped us figure out how to build a better model.

I’ve been proving my value through conversations, through the points I make, through the work I show specific people. But I need to make it more evident. That’s what this entire journey is about – filling the gaps without being prey to anyone else’s belief system or narrative.

I want to run this show. I want to be good at math, at doing math, at applying it, at solving problems. Maybe I’m becoming a super generalist, but data science itself is already a pretty specific field. The coursework demands foundational machine learning understanding, while continuous engagement with people makes me think about helping them solve new, varied problem statements.

The Reality of Growth

Professor Elias’s causal inference class makes me think deeply – not just learning methodology, but learning new paradigms. I see problems through the lens of what I learn there, and it improves my thinking about what problem statements to tackle. Maybe in upcoming posts, I should share more details about the different problem statements and approaches I’m exploring.

Sometimes people don’t get you, and that’s fine. Everyone has their own judgments. I might be completely wrong sometimes, but I’m quite retrospective – I always look back and try to fix things that are possible to fix.

Our ideas might not be acceptable to peers. People might feel they don’t conform. But that’s fine too. Someday I’ll be in a position where people find me right. For now, I’m putting all my energy into building things, learning foundations, learning the right things.

I’m not going to stop learning, but I am going to speed up my efforts into building something solid. Things might take time to become evident and visible, but we’re on a journey that makes it possible.

The Bottom Line

This reflection is going to be just another reflection, a small self-talk with myself where I can address the pros and cons, where I can see where I need to build more solid efforts and where I should leverage what I’ve built and done – the learnings I’ve accumulated, the experience I’ve accumulated.

This isn’t for anyone else – maybe people will find it relevant, but it’s truly for myself. I shouldn’t feel ashamed or feel down that I haven’t done anything. I’ve come far, hit milestones, built trust, created valuable impact.

Yes, there are things I missed building that I pledge to build in the upcoming days. And I will build them. I’m retrospective and quick to fix what’s possible.

I should feel proud. I shouldn’t change my behavior for the sake of others, but should change the way I work in this ecosystem by truly listening to my heart and building what I wish to do – with more focus, more consistency, better management of what’s on my plate.

But I won’t stop sharing my learning or seeking new learnings or working with people who need my support. Because that’s who I am, and that foundation is worth building on.

Don’t feel low. Keep going. Keep doing things.