Ansh Tandon

Thrilled to publicly share that cova.care is a part of YCS26. Home care will be the front line of how our ageing America gets cared for. We’re now licensed and live as a provider in Michigan.

@viriditax @iskander interesting, but IRBs would still have to parse through info…and are trial designs really standardisable? don’t have much expertise in this domain so would love your thoughts

@anshtdn @iskander idea is templates help them assess the trials & data requests much faster by flagging exceptions, e.g., "template 2: Ph0 PD response assay in healthy volunteers @ 10% NOAEL dose" vs "template 7: Ph1 3x3 dose escalation in {inclusion criteria x}, add CA-125 quantification"

Let’s dream a bit. How would you dramatically reorg biotech research / drug development to make dramatically faster progress towards curative therapies? (Don’t say “use AI”, hypothesis/candidate generation under current structure isn’t a bottleneck)

@viriditax @iskander to your second point, why templates? i think the bottleneck to IRB approvals is the human element of gathering clinicians to assess trials and data requests. what about this idea: IRB copilots that flag ethical nuances and help boards make decisions quicker?

@iskander + FDA responses in <90 days + template approved trial designs with <10 day IRB approval + drop non-rodent safety/tox req for Ph1 trials in terminal patients + fungible accelerated review credits + more flexible post-trial stat stratification for approval on responder attributes

benchmarking cell segmentation in spatial transcriptomics is so dogshit. we’re still relying on pixel-level accuracy metrics like IoU and NOT shifting the evaluation framework towards biological meaningfulness. for starters why don’t we use silhouette scores on UMAPs?

@brianyunmd pattern recognition is just one part—what about empathy, ethics, and the messy realities of real patients? aren’t we overselling what algorithms can actually do?

as a MD and as an engineer, this is a topic I've thought long about. and this post is only half correct. hospitals serve two purposes: knowledge center and a treatment center. medicine as an academia is a pattern recognition of billions of deaths and illnesses throughout human history. there is less logic involved than people think, more about memory and data collection (what to measure, when to measure, and how to interpret). in this function, I am certain that AI will replace 90% of doctors in a short period of time. you can even argue that the tipping point has already happened. however, actually giving out the treatment and the care is another major function of medicine - the 'problem solving' and 'get shit done' layer of the job. drinking water in this case was a very accessible form of treatment which is great, but in most cases treatment is delivered through pills or the hands of an expert. this part of the doctor's job is not really replaceable by AI until a robotics breakthrough happens. i do want this change to happen because millions of lives are being lost due to the space time accessibility problem of healthcare. in fact, this is a problem I have vouched to solve as an entrepreneur since I was 16. it was the whole reason I went to med school and learned how to code. we are working day and night to bring this future closer to humanity. what an awesome version of the world would that be. I've written a whole blog post about the nature of medicine: blog.brianyunmd.com/p/what-wouldnt… , and the future we want to build: blog.brianyunmd.com/p/the-future-i…

@EQL_AI mums just spent 30mins conversing with your triage chatbot that “replicates human interaction with a personal touch”. if healthcare's resorting to crappy ai tools that regurgitate the same y/n qs then fuck me, we're further away from patient-centred care than i thought 😂

imagine if we could optimise blood glucose patches to have feedback loops so that dose delivery could be adjusted in real time based on activity intensity. connect to whoop perhaps? super useful for athletes

@kasratweets may relate to the balance of adhesive and cohesive forces, which affect tissue packing and cell behavior

the big question here is: why does this happen around 300 (±100) cells? how do the cells "know" they've reached that number? it's not clear. one proposal is that it's related to the optimal length scale for the biochemical reactions involved in symmetry breaking & polarization

this is one of the cooler biology results I've seen. a group of mouse stem cells in a petri dish can differentiate into any of the cells needed for an embryo, but they always create a very disorganized, incoherent embryo. that is, unless you have exactly 300 cells:

@NikoMcCarty cell-free protein synthesis: ditch cells. use optimised lysates. higher yields, less purification hassle. scale up with automation (robotics?)

Reducing costs for protein synthesis still seems like a highly underrated problem in biotech. An average-sized human protein costs >$100 to make; gene synthesis is far too expensive. Then you have to purify, etc. Direct protein synthesis is inefficient and limited to ~150 amino acids, right? And it's still difficult to make large amounts. This seems like a major bottleneck to scale protein design. How do we solve it?

@WillManidis great take. i get the idea of patients taking action (thus the dashboards) but humans are simple, they just want to “feel” good. im a fan of holistic med, esp since ive seen it help someone close with autoimmune issues. listening is something healthcare needs to adopt universally

Every week, a dozen new companies boasting "tech-enabled healthcare for the rich" announce massive funding rounds. I think nearly every single one fundamentally misunderstands the business of high end medical care. Here's how to build a real executive physical killer:

@timhwang An Integrated Brain-Machine Interface Platform With Thousands of Channels

has a white paper ever changed the world? which ones?

@IR5Iconoclast absolutely, and yes, it's a step. but to an extent feels like a gimmick. it's not hard to develop a more interactive interface which could enhance ux beyond just notification pop-ups.

I don't think technology will ever replace human friendship, but something like this could really help people who do not have anyone in their lives at a point in time. Particularly if it is designed to support mental health. It might even be able to help people form human friendships.

Is 'Friend' the next biggest thing? Loneliness is a rising public health issue, with almost a quarter of the world feeling lonely, according to a Meta-Gallup survey of more than 140 countries. Loneliness is greatest among young adults, aged 19 to 29. Friend could make a big difference to our mental health, with recent research demonstrating that social support from artificial intelligence is beneficial.

@DiogoMCaetano @LawyerLiam was literally thinking about this the other day. currently have to convert ableton tracks into .wav files to collab. can’t send ableton files because the recipient might not have the same presets + vsts. the wavs end up being too large to send over email. absolute hassle

@LawyerLiam The collaborative music software. All in your browser. jamminstudio.co

Startups - What are you working on? - Pitch in 10 words max: - Website URL: Let's go 🚀

@jacobrintamaki lasik was once considered extreme, now it’s pretty routine. give it time, it’ll get more noninvasive as well. and tbh I don’t think their goal is to reach billions, just ppl that perceive its value to be > than the risk.

Neuralink, as cool as it is, will fail because it requires cutting people’s skulls open. Any bidirectional neural interface technology must not cut people’s skulls open if you want to reach billions.

@sidddddesh berkshirehathaway.com

@levelsio disagree lol. then you’ll be stuck in the rabbit hole of over-engineering a solution that no one rlly needs.

Still the most common mistake by software developers who switch to building startups I think Focus on tech, not on getting customers

@businessbarista @shl @ezrainc nice tech. full-body cancer scans but 'lungs not included' is a bit crazy though. pay-to-win. i get the fixed-price model for transparency but hope they can get insurance coverage for LIFs

This founder is pitching this world-class investor (@shl) to take a follow-up meeting and invest $100,000 in his business. Will Sahil say yes? The Business: @ezrainc The Problem: People die of cancer. The Solution: Ezra helps people find cancer early using a full body MRI powered by a number of AIs. Would you take a meeting to invest? Comment below👇

@t_blom @LenireTinnitus Oto Health (YC W'22) - joinoto.com worked with them as a data analyst back when i was in high school. they've built digital therapeutics (like audio masks based on cbt). would definitely take a look.

I have had bad tinnitus for the last 3 years. I scored around 75 on THI - it was pretty debilitating. I tried cutting out alcohol, caffeine, sleeping more - nothing really helped. For the last three months, I've used a new FDA-approved device from @LenireTinnitus. I now score about 15-20. The tinnitus is barely noticable. It's not cheap (around $4000) but it has improved my life pretty dramatically. nature.com/articles/s4159…

@FitFounder habit stacking like going on a walk + facetiming mates

What's an underrated health habit you'd added to your life that's made it better?