Starting Over, on Purpose
I’ve been meaning to write more.
Not because the internet needs another blog — but because writing is how I think. And clear thinking is the bottleneck in everything I build.
This site is a reset. Not just a visual one — a structural one.
The previous version reflected a different chapter: deep dives into Scala, Spark, and distributed data systems. That work shaped me. But the questions I’m working on now are different.
This is where I am today.
What Changed
For years, I operated at the infrastructure layer — pipelines, lakehouse architectures, distributed compute, performance tuning.
The goal was always the same: deterministic transformation and transportation of data from one end to the other. You defined the logic, the pipeline executed it exactly as programmed, and when it failed, the failure was predictable. A schema mismatch. A resource limit. A partition skew. You could debug it because the system did what you told it to do — nothing more, nothing less.
Then something shifted.
I started asking a different question: what if these pipelines had intelligence built into them? Not just the ability to move and transform data, but the ability to interpret it — to notice anomalies, make routing decisions, recover from ambiguity without a human writing the recovery logic first.
That question pulled me out of the infrastructure layer and into something new.
Software doesn’t just execute instructions anymore — it exhibits behavior.
The frontier is no longer about moving data efficiently. It’s about systems that interpret context, make decisions, and collaborate with humans. The shift isn’t that systems are becoming more complex — it’s that their outputs are no longer fully predictable from their inputs.
That changes everything.
The architectural challenges are new:
- How do you design autonomy without chaos?
- Where should humans remain in control?
- What does reliability mean when outputs are probabilistic?
These are no longer theoretical questions.
They’re product questions.
What I’ll Be Writing About
This site is my public working notebook — structured thinking, not trend commentary.
Three areas keep pulling me back.
Designing Agentic Systems for Production
Not demos. Not toy agents. Systems that survive ambiguity, failure modes, and real users. I want to write about orchestration patterns, guardrails, evaluation frameworks, and what it actually means to earn operational trust when your system’s outputs are probabilistic. Failure isn’t something to eliminate — it’s a measurable hypothesis.
The Interaction Layer Between Humans and Agents
I’m especially interested in how humans and AI systems interact.
Most tools either overwhelm us with metrics or reduce everything to chat. What’s missing is guidance — interfaces designed around the next best action.
If we get that wrong, autonomy becomes noise. If we get it right, AI becomes leverage.
Thinking Like a Builder, Not a Feature Shipper
The longer I build, the more I’m drawn to foundational questions: How do you identify the right problem? What actually defines an MVP in an AI-native world? How do you make disciplined build-versus-buy decisions? When is a system “good enough” to ship?
AI dramatically accelerates execution. It does not eliminate the need for judgment.
In fact, it raises the cost of poor judgment. AI doesn’t just lower the cost of building — it lowers the cost of building the wrong thing convincingly. A mediocre feature shipped fast with AI assistance looks polished enough to survive review. When building becomes easier, deciding what not to build becomes the real constraint.
Why Rebuild the Site
The tooling makes rebuilding easier than ever.
But speed isn’t the point.
I rebuilt this site so the publishing workflow itself reflects how I think: idea, hypothesis, pressure test, refine, publish. Fewer posts, higher signal. Arguments that survive scrutiny.
If I can’t defend the idea rigorously, it doesn’t ship.
Why “gcdaii”
It’s my initials plus a small personal detail.
In Nepali, dai means “older brother.” Friends have called me “gcdai” for years. I added the extra “i” simply because the handle was available.
If you’re working at the intersection of data, AI, and product — especially if you care about signal more than noise — I’d love to connect.
We’re early in this shift. Let’s build deliberately.
Learning Never Stops.
continue reading
Your Code Survives the Session. Your Reasoning Doesn't.
A practical guide to building a personal journaling system for AI-assisted development — using Claude Code skills and hooks to capture the reasoning that code alone can't preserve.
Building the AI-Ready Data Platform
Agents don't fail because the model is wrong. They fail because the data platform wasn't built for them. Here's what an AI-ready data layer actually looks like — and proof it works.
Data Is the Bottleneck in the Agentic Era
Good data yields good agents — and agents can produce far more decision data than humans ever did. The data layer is about to become the most critical piece of the stack.