Arindam Majumder

Everyone's building personal agents. But they're solving the wrong problem.

We've seen it with OpenClaw going viral, Hermes agents, all these harnesses popping up that let you run your own agent.

But everyone is missing one very important thing!

The harnesses are solid. The models aren't personalized. That's the actual bottleneck.

Every person uses agents differently. Different tasks, different workflows, different preferences. Your taste matters here.

If you want a truly personal agent, the harness is only half of it. The LLM itself needs to be tuned to how YOU work.

I think we're headed toward a future where personalized LLMs become the norm. Not just generic models with fancy wrappers.

Open-source models are closing the gap fast. The infrastructure is there.

So, how do you personalize your model?

Your inference logs.

That's it. Every interaction you have with the model, every way you prompt it, every task you run. It's all sitting there in your logs.

You can take that data and fine-tune directly on it.

This makes building a real personal agent way more practical than people think.

1 week ago | [YT] | 7

Arindam Majumder

My Cursor Agents SDK video is already ranking on Google.

Working on a deeper breakdown now.

Stay tuned.

2 weeks ago | [YT] | 5

Arindam Majumder

Last week I spoke at GDG Kolkata's Build with AI event.

First time speaking at any GDG event. It was special.

I was nervous before going on stage. Kept thinking about what could go wrong. But the moment I started talking, everything just flowed.

I walked through building a real-time voice agent with Gemini 3.1 Flash and did a live demo. People were genuinely excited. You could feel the energy in the room.

That reaction made it all worth it. It's one thing to build stuff and share it online. It's completely different when you see people light up about the possibilities in person.

Big thanks to the GDG Kolkata team for giving me this opportunity.

Looking forward to doing more of these.

Skip the Carousel

1 2 3

2 weeks ago | [YT] | 7

Arindam Majumder

Just crossed 100k Views on my YouTube channel 🤯

Honestly, this means a lot to me.

I started making videos just to share things I was learning and building, and seeing so many people watch, support, comment, and stick around is genuinely crazy.

Huge thanks to everyone who supported the channel till now ❤️

And we’re just getting started 👀 Got a lot of new videos, experiments, and deep-dives coming soon.

Stay tuned!

2 weeks ago | [YT] | 6

Arindam Majumder

Claude Opus 4.7 has a new tokenizer.

Anthropic says it uses 1.0-1.35× more tokens.

But real tests with developer prompts? ~1.47× more.

What this means:

• context fills faster
• sessions end earlier
• API costs go up

Went through the details and broke it down here 👇

1 month ago | [YT] | 1

Arindam Majumder

Just hit 2k subscribers on YouTube 🎉

Not a huge milestone, but it means a lot. Thanks for supporting the content.

Gonna be way more consistent now.

1 month ago | [YT] | 10

Arindam Majumder

I had no idea how much I was spending on Claude Code.

Hundreds of sessions. Thousands of tool calls. Cache hits? No clue.

So, I built cc-lens to see what's actually happening.

github.com/Arindam200/cc-lens

Cost by project. Cost by model. Cache efficiency. Everything Claude Code does, visualized.

1 month ago | [YT] | 4

Arindam Majumder

Claude Code bills you but tells you nothing.

Introducing CC-lens.

Your sessions. Your spending. Your cache efficiency, laid out in one dashboard.

See exactly where your tokens go. Which projects burn budget. What actually caches.

Just Run: npx cc-lens

1 month ago | [YT] | 5

Arindam Majumder

Working on this Guide!

Will share a detailed Tutorial around this.

Stay Tuned!

5 months ago | [YT] | 11

Arindam Majumder

Started Fine-tuning Models on ‪@nebiusofficial‬

Will share a Detailed Guide Soon!

6 months ago | [YT] | 9