samit

🇮🇳🇨🇳🇨🇦 Founder prev. company acq.

[samit @samitkhalsa]

i left notebooklm to solve a real problem in education.

@ Penseum we built the opposite.

no waiting — start learning in seconds

no walls of text — your tutor talks, visuals appear as it explains

no wasted time — flows with your understanding, deeper where you're stuck

gets smarter — every session knows you better than the last

@penseum_'s ai tutor. live now to our million+ users.

(i meant i left notebooklm as a user)
@KyushiThe

[samit @samitkhalsa]

6/ My progress so far:

Down from 237 → 220 lbs, my lowest in a year.

Fat percent 31% -> 29.2%

Will be taking another blood test to see how my biomarkers have moved in 1 month

[samit @samitkhalsa]

1/ every health app sucks. whoop, apple health, all of them. numbers with zero context.

I was measuring everything and still felt like shit.

so i built my own health brain. used @karpathy methods to create an AI clone of me that knows everything, injuries, diet, full history.

i call it Doppel

hooked it up to my weighing scale, whoop and uploaded all my history, every blood panel piped into one system that reasons across all of it like a doctor who never forgets a data point.

@bryan_johnson waddya think?

[Andrej Karpathy @karpathy]

LLM Knowledge Bases

Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So:

Data ingest:
I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them.

IDE:
I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides).

Q&A:
Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale.

Output:
Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base.

Linting:
I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into.

Extra tools:
I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries.

Further explorations:
As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows.

TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.

[samit @samitkhalsa]

5/ I built it on myself.

Now I'm opening it up. If you've got wearable data, bloodwork, or a DEXA scan and want your own AI body clone reading it all, here's the skeleton of my code

Repo👇
https://t.co/tkJNWJjFbT

[samit @samitkhalsa]

@chamath i am.

[samit @samitkhalsa]

We might not have a launch video like this, but we’ve delivered our AI tutor to 1.6 million students in less than 5 months.

No waitlist, no hype…. just real learning at scale.

This just gives me more confidence that we’re ahead and building faster than anyone 😮‍💨

[Sue @suekhim]

AI is making kids dumber.

It should be making them geniuses.

Introducing Koji, the first AI tutor that gets kids to actually think. 👇

[Aaron Paul @_aaronpaul25]

This situation is actually very simple.

The issue is not whether Nicole ever mentioned me. She did mention me.

The issue is that mentioning someone is not the same as accurately crediting them.

From the outside, the podcast framing makes it sound like she was the solo founder/operator behind Glam Up and Sprout, and that everyone else was basically a footnote.

That is not accurate.

Does this situation seem petty & insignificant from the outside? Probably tbh.

But when you put your heart and soul into building something, credit matters. Not just to me, but to the team and Nicole too.

So yes, the record being accurate matters.

And so to be very clear, I did not “just do the dev work.”

In the interview, she says her account was the first to go viral. That is not true. The first account/post to go viral for Glam Up was the one I ran, and that account ended up accounting for roughly 40% of the views.

Similar situation with Sprout. A creator doing the faceless “war is over” content went viral first. Not her. She may have been the first to go viral with on-camera content, but she was not the first to go viral for the company.

I designed the actual Glam Up and Sprout apps end-to-end, excluding paywall videos.

I built the apps, funded/incorporated the companies, held majority ownership, recruited the team, and helped drive early growth alongside other cofounders and team members.

And this is not just about me.

Dillon deserves serious credit too. On the UGC/growth side, he helped build the bulk of the engine and has continued operating, testing, and improving it long after Nicole left in September 2025.

Sprout did not freeze in time when she left. The team kept building.

So when the story gets framed as one person running the show, it does not just sideline me. It sidelines Dillon and a lot of other people who materially contributed too.

Nicole contributed meaningfully to UGC/growth. I’ve never denied that.

This is also not about saying minority ownership means someone contributed nothing. That is not my point.

The point is that ownership, product, design, funding, building, and team contributions should be represented accurately.

But “I do everything but coding” is not true. Not even close. That is something she explicitly said.

On the Starter Story point: yes, I declined to go on it and requested she not do it. She asked me multiple times and I requested multiple times for her not to do it, but I eventually agreed. I was only told about Starter Story, not Superwall.

Either way, the issue was never simply “there was a podcast.” The issue was the overall framing once the interviews came out.

A later shoutout does not fix the core framing of the interview itself.

It’s also strange to say I’m building a public brand off her name when the whole reason this started is that she is building public content around apps I designed, built, owned, funded, and am still actively working on.

On the consulting point: I’ve said multiple times that 100% of any consulting money from this will go to charity. That is thoroughly documented, and once the calls are done, I’ll share the donation receipts.

I don’t want to personally profit from drama around apps my team is still operating.

I’m not trying to sell a course or become a coaching guru off this. I’m trying to correct the record.

She also has another app doing well. She could talk about that. Instead, the public content is largely around apps my team and I are still operating, which shifts copycat risk onto us while protecting her own current thing.

The implication that my UGC playbook was repurposed from her thesis is false.

Her thesis is a founder-level case study. My playbook is an operator-level creator program guide based on what we actually tested, ran, and scaled.

They overlap on some principles because they are about the same business. That does not make them copied.

Also, I’m not sure why the internal UGC course is being framed as a point against Sprout.

I’m not denying that Nicole recorded and contributed to creator training materials while she was working with the company. She did.

But that course was created for the company, used by the company, and refined with input from the team. Dillon, other team members, and I contributed feedback and operational knowledge from actually running the program: recruiting creators, coaching them, testing formats, improving scripts, and scaling what worked.

She was compensated for that work and still retains equity in the company.

My point is much narrower: my public UGC playbook is not a repackaged version of her thesis.

I’m human, not an LLM. I got angry and said things privately that I regret. I should have handled parts of this better.

But anything I said in anger came from feeling betrayed and exploited over the way my work, my team’s work, and our contributions were being represented. It was about credit, contribution, and character.

That does not mean I handled every message well. I didn’t.

But it also does not mean the public framing is accurate, or that this was some random one-sided harassment campaign.

I have screenshots too but I’m just trying not to turn this into a screenshot war.

These were not solo founder stories and the record should accurately reflect that.

[Aaron Paul @_aaronpaul25]

Second time I’ve seen @madebycol tell the Glam Up / Sprout story in a way that materially misrepresents who built what.

She contributed. But this was not a solo founder story, and I’m not going to keep letting my work, nor my team's work get erased.

It's frustrating, sad & disappointing.

[Aaron Paul @_aaronpaul25]

Cool to see this story getting attention. A short interview naturally compresses a lot of context, so I’ll add my side.

I was Co-founder & CEO across Glam Up and Sprout, led much of the product, growth, and execution behind both, and owned all the design + engineering for Glam Up.

Glam Up reached 1M+ downloads in under 6 months, fully bootstrapped. Sprout is now bootstrapped to $250k+ MRR.

The highlight reel is fun to talk about. The operating system behind it is what’s actually valuable.

I’ve stayed mostly quiet online, but I’m going to start sharing the playbook behind building AI mobile + web apps from 0 → scale.

Verified metrics below 👇

[samit @samitkhalsa]

@tszzl Been a chindian since 1998

[samit @samitkhalsa]

@andruyeung Lunch lady

[samit @samitkhalsa]

@itsjessyin North of brooklyns better imo

[samit @samitkhalsa]

@alexinbinary @polarityco lfg 🚀

[Alex @alexinbinary]

At @polarityco we're giving teams the validation they need to just do things.

Announcing project Keystone & Paragon.

[samit @samitkhalsa]

nobody warned me about this part of SF.

you walk into a room and everyone assumes you can build something massive. like it's a given. that energy is wild.

but the flip side? they also assume you can lose it all tomorrow. 0 to 1 or 1 to 0, any day. no in between.

[samit @samitkhalsa]

@vitaliidodonov Lfg

[samit @samitkhalsa]

@SpaceX @cursor_ai dafuq

[samit @samitkhalsa]

@MessiahYaniss16 🥲 pls no