21 | Won 100k cloud funding from Google | Beat the Anthropic open recruitment challenge | Stanford ASES fellow | @join_ef
Joined April 2022
- Tweets 874
- Following 1,235
- Followers 1,368
- Likes 755
108 Photos and videos
Pinned Tweet
May 10
I will defeat death and achieve biological and absolute freedom
1
7
1,193
Jun 6
Honored to win 1st place at South Asia’s biggest AI conference and be among the three teams invited to present our work.
We built an AI Co-Scientist for Biotech Researchers that beat the Industry leader by 40% and a Virtual Cell model that is currently top 5 globally in benchmarks.
None of this would have been possible without the mentorship of @lmthang and @HoangDuon, and I am glad to have them as my advisors.
9
1
39
1,850
May 30
One of the most awesome person I have met!
May 30
The day I missed my bachelor's convocation ceremony was also the day I got into Entrepreneurs First (@join_ef )
Sometimes the next chapter starts before you've had time to celebrate the last one.
At EF, we're exploring the intersection of Biology × AI × Computing.
Our moonshot: accelerating scientific discovery through better computation and AI.
Excited for what comes next. ✨
2
3
26
2,854
May 30
I am in Vietnam for a couple of meetings with the local biotech companies,
would love to meet folks interested in biotech, AI drug discovery, autonomous Labs, AI and robotics in general, comment or send me a DM.
We are also going to Singapore and Japan next so would love to meet folks from there too.
5
1
59
3,275
May 24
Thinking of starting a community for people who want to learn biotech, drug discovery or anything bio in general.
Reply to the tweet if you want in.
Let's kill Death together
85
16
203
11,090
May 16
This is possibly the most inspiring picture ever.
2
8
729
May 13
Just watched @parmita's podcast - always good to find more longevity believers, there should be more sane people in the world.
Here are some thoughts from a noob:
1) Her point on CRISPR was brutal but spot-on: tweak one gene for “higher intelligence” and suddenly your lungs are different. Single edits don’t work in complex systems.
but this reminds me of @demishassabis's talk with @garrytan where he advised to hunt areas with insane combinatorial spaces, a clear success metric, and a way in via data or simulators.
maybe AI could be the unlock here?
2) I'd like to push back on her point about LLMs being good at code because all the code already exists on a dataset. I think it's more because of verifiable rewards. the game changer might be a way to setup good, fast feedback loops….imagine agents actually running labs and equipment 24/7.
3) I like her points about how a company should ideally have some IP involved or regulation unlock involved, hits especially hard bcoz we seem to be stuck in pivot hell lol
loved the quote "Every day you wake up and you're blinded by your mission. You are not married to a technique", I might end up stealing this. ^_^
thanks @b1shtream for recommending this to me.
1
2
26
5,214
May 5
The cruelest thing about job hunting aren't the rejections,
It's that, you will never know why you got rejected.
Well here's Donna, She gives you
1) specific feedback on why you got rejected
2) a side by side comparision with the candidates who actually got shortlisted.
try it out!
4
16
1,171
Apr 30
The best advice for people building in Deeptech or AI for Science from the @ycombinator X @demishassabis podcast (based on AlphaGo and AlphaFold)
Look for problems and fields which have
1) a massive combinatorial search space
2) where you can define success as an objective function
3) initial entry point through either a dataset or a simulator
and sit with it for 10 years. take into account that AGI will arrive somewhere in that timeline and account for it and go wild.
PS: thanks @garrytan for hosting the pod, very thought-provoking
11
4
123
7,706
Apr 30
Just watched the one-hour podcast of Legora's CEO so you don't have to
1) Every body is already using ChatGPT and Claude daily. To make them switch to your product, you need to be three standard deviations better. No half measures. That means obsessing over model capabilities.
2) You must be willing to kill things you spent 6 months building. The AI landscape moves so fast that clinging to old work slows you down.
3) Deeply understand your users. The founders used to pay for lawyers’ lunches just to learn how IP law actually works. That domain insight became their biggest edge.
4) Every executive (and honestly every employee) has an expiration date. The company evolves so fast that someone who was perfect 6 months ago might not fit the “new” company.
5) Evals are everything. Their proprietary evaluation framework gave them clarity and speed. Development accelerates massively with good feedback loops, Without good evals, you can’t move fast with conviction.
6) Product quality wins. The goal with any product is that after 30 days of pilot, if you shut it down, the employees of your client should riot.
7) There’s almost no point in making a long-term roadmap. You need to react to the market constantly and stay extremely adaptable.
8) Ruthless focus: Instead of building 15 average features, they obsessed over getting just 3 things exceptionally right.
Awesome podcast by @MaxJunestrand, @jaltma and @chetanp
8
522
Apr 25
I am actually in disbelief about how much work garry tan is able to get done.
Opened a PR on gbrain last week about an incremental sync deadlock fix.
I remembered to check a week later while he had already shipped it in 2 days, all while managing YC. How??
3
684
Apr 21
I just have an active claude code session on my VPS connected to my mobile and have it debug anytime my openclaw breaks
Apr 21
I asked this exact question to a colleague today.
Me: I really can't get to grips with OpenClaw. It often takes 1hr to reply to simple messages and often just dies randomly
Him: Do you have a Claude Code session monitoring it?
Me: no...
Him: Yeh it is impossible to use it without CC babysitting lol
2
3
636
Apr 20
Claude code should have a revenue sharing program for any community made skills.
Things like gstack pretty heavily change the experience, so an incentive program to encourage similar innovation makes sense.
3
302
Apr 18
Genuinely cool take,
Hats off to YC for organizing this super event, I am sure it must not have been easy.
Apr 18
everyone’s always got time to hate on things but somehow no one shows up when it’s time to appreciate something genuinely good
so i’ll do that today
Y Combinator just hosted their first ever startup school in india. in bangalore.
25,000 people applied. they picked 2000.
@snowmaker flew in from SF.
@harshilmathur @lkeshre @aadit_palicha @viditaatrey all YC founders. many yc partners sitting in electronic city on a saturday, just looking for the next great company from india. betting on young builders.
and barely anyone’s talking about it?
we’ll spend days quoting bad takes. pile on someone for a typo. write threads on why india can’t build.
but when something like this happens silence. or worse, trolling.
there was a 14-year-old hosting the pre-party
a kid from patna building ai video tools
a 17-year-old from jehanabad building cancer screening tools for asha workers after losing his aunt
aligned with management could’ve been better. selection wasn’t perfect. [food and venue valid]
here’s what i actually think about the selection:
most people in that room were college and high school students. the ones with strong profiles have already figured it. they’re on their path, they’ll be fine.
but the young kid who just got in? who sat in a room with actual YC founders for the first time? that changes something in them.
you don’t need to teach young people how to build. you need to sell them the dream that they can. watching someone who came from nothing and made something meaningful
that’s the whole thing. that’s what shifts identity.
and i want to see more women here too
more women in STEM. more women building. more women who can change the world
that’s something i’m actively trying to fix with @aiweekendsxyz too.
hate is easy. building is hard.
and apparently appreciating people who are building is even harder.
yes the selection had issues. yes things could be run better. but we should
appreciate the parts that deserve appreciation, not just troll the whole thing because it wasn’t perfect.
@ycombinator hope this is the first of many.
the indian startup scene just got a pretty loud signal. most people were just too busy arguing to notice.
🇮🇳 /acc
9
705
Apr 16
is it just me or is Opus 4.7 10x Worse than 4.6 for everyone?
It doesn't use any tools. It doesn't check out like web search in depth. It doesn't go the extra mile, unless I force it to.
Is Opus 4.7 actually just a scam by Anthropic to reduce their compute demands?
3
315
Apr 16
Just saw "the last lecture" by Randy Pausch and I haven't bawled like this in months lol,
Truly the greatest head fake of all time
1
195