Benjamin Lee

@fuma_nama @schanuelmiller What does {$} do that $[.]md.ts doesn’t? Haven’t seen anything in the Router or Start docs about {$}

@schanuelmiller Cool just learned the difference between $ and {$} in file names, it's also possible in Tanstack Start with {$}[.]md.ts

I am really impressed by Waku.js, all other React.js frameworks needed middleware + redirect. Waku just works, both "/docs" & "/docs.webp" renders correctly, no conflicts.

@schanuelmiller @kokaneka @tannerlinsley @tan_stack Is there a world where server functions can have OpenAPI specs? I know Nitro 3 just added beta support (didn’t work with Start when I tested) and there’s always oRPC and Elysia but it would be great to have out of the box

@kokaneka @tannerlinsley @tan_stack server functions have special serialization that indeed is not that readable. this is necessary to allow serializing much more than what pure JSON supports (e.g. Promise, streams). we are building out devtools for start so you can inspect server function calls and responses there

Hey @tannerlinsley I'm trying out @tan_stack for a personal project. Looking at the API requests in the network tab for debugging purposes is one of my go-to techniques for debugging. Do I need to let go of it if I want to move to server functions? It looks like this rn

@AlemTuzlak @tan_stack @thvdnta Ladies and gentlemen, we got him Welcome to the team @thvdnta and thanks @tannerlinsley and @AlemTuzlak for the intro!

A good friend and a contributor to @tan_stack is looking for an internship opportunity. 👀 If your looking for amazing devs as interns who are enthusiastic and love the craft DM @thvdnta! 🚀

@Rippling has the most infuriating sales process I've ever seen. I've been bounced between four reps and had three meetings scheduled and then canceled by them... and I still haven't gotten pricing!

@battjonesuk @realdocspeaks Not affiliated with them (never used them myself either) but Radiology Assist has self pay MRI pricing and availability listings. For an upper extremity joint MRI in my hometown, it’s less than $500 all in. @realdocspeaks is essentially right on the out of pocket pricing

@realdocspeaks I’ve never seen an MRI below $1000 and they average at like $2000 where I’m at, in one of the poorest places in America. I wonder if that’s bc of Medicaid.

You are looking at this the wrong way. MRIs are too cheap to insure, and patients should pay out of pocket for them. In my ZIP code, an MRI of the lumbar spine costs $244.85. Why would we insure something that costs less than $300? A maintenance package on a Honda that includes an oil change, tire rotation, brake inspection, and filter replacement costs the same as the MRI. Insurance should be reserved for catastrophic events.

@GeneralChaos42 @realdocspeaks Yes but there isn’t an infinite amount of them. There are only so many MRI scans that can be done in a day. There must be some mechanism for allocating these scans. It would be best if they were used on people who could benefit from them.

@Benjamin_D_Lee @realdocspeaks More than one MRI machine exists…

The real move would be to eliminate prior authorizations!! We don't need more "tech solutions," we need insurance solutions. We need to minimize the role of insurance and remove it from: • labs • imaging • prescription drugs • physician services • procedures

I agree with this. Insurance shouldn’t be in the business of deciding whether a $300 procedure is necessarily for the same reason they shouldn’t be paying. But what about much more expensive procedures? It’s the same problem again. A few bad actors can force PAs on everyone.

@nurees1987 @ONC_HealthIT @mcuban Faxes are actually much better than portals for just this reason. 100% interoperability and backwards compatibility. Multimodal AI agents can handle faxes as inputs with no problem but you can always fall back to a human easily.

@ONC_HealthIT @mcuban Multiple point solutions already exist in the market that support this operability including the use of agentic AI. The issue is adoption. Practices don't have time to utilize yet more portals. Slap a page in the fax and done.

Some doctors’ offices still print records, fax them to health plans, and wait for a response. That’s exactly the kind of friction health IT should eliminate. New standards can help systems talk directly to each other, reducing burden and moving prior authorization closer to real time. #Interoperability #PriorAuthorization @mcuban

@DougAlfordMD1 @realdocspeaks It is ridiculous but automation can help immensely. There’s no reason why a human has to sit on hold or copy data into a payer portal. We have AI doing this for MRIs on a daily basis and it works exceptionally well.

All day long prior authorizations and pre-certifications. Having to have a full-time employee in primary care just to do prior authorizations and pre-certifications is totally insane and a waste of money. This has got to end at some point. I don’t own an MRI and I have no financial incentive from prescribing the best medicines. Please help primary care physicians with this burden.

@w_milczynska @doctorbhargav The privacy/compliance isn’t an add on though. OpenAI isn’t disclosing patient information regardless of whether a HIPAA BAA is in place. Not every doctor can sign a BAA depending on where they work but for those who can, they can fully comply with HIPAA.

@doctorbhargav agreed, spot on list! just finding it v weird that they made HIPAA compliance an optional part. you can't build for healthcare and treat patient privacy as an add-on

OpenAI just made ChatGPT free for clinicians. Most people will see this as a product update. They’re missing the real shift. AI isn’t replacing doctors. It’s taking over the highest-friction work: * Documentation * Literature search * Prior auth * Patient comms That’s how adoption actually happens.

@DrDiGiorgio A lot of the copy-pasting would be fixable if EHRs weren’t blocking write access from third parties. The 21st Century Cures Act only mandated read-access, so chart review automation is doable (my startup does this) but no luck for getting rid of copy-pasting for now

So much AI potential is falling short in healthcare. Hospital admin are forcing doctors to copy-paste a clinic H&P into a new note for surgery instead of automating it. Hospitals are paying NPs mid-6-figure salaries to do chart review instead of automating it.

@BryanOnel86 Is it really true that @oneleet has only done SOC 2 Type 1? That's got to be a mistake in the trust center

@MichaelAlbertMD Voice agents are uniquely suited to that population. I’ve deployed them in the retina space (where patients are older and vision impaired) and it worked really well. They just call up the number and talk on the phone

All of these ACCESS programs are built around AI and digital-first workflows—for Medicare beneficiaries. It makes you wonder: did anyone designing this actually spend time caring for older adults? Many of these people still use flip phones. They’re not engaging with AI chatbots. They’re not self-scheduling online. In our clinic, staff routinely call them just to get appointments set up. There’s a real risk here: if the model assumes digital fluency that doesn’t exist, it won’t just underperform—it may fail at scale. Will this become one of the biggest healthcare misses we’ve seen?

@pimentcrypto @alc2022 The US government can regulate peptides, even if you synthesize them yourself for your own use.

@Benjamin_D_Lee @alc2022 Can you explain

peptides can’t be regulated. trying to is like regulating code. 100 million can be synthesized. the value is in the AI that knows which one you need. that system is a digital twin of your body. this is Palantir at $7. same thesis. same upside.

@aa22396584 @MaziyarPanahi Assuming this is an AI slop comment (em dash) because that's not how HIPAA works. There are no HIPAA workloads that this touches that couldn't already use the cloud because all major cloud providers will do HIPAA BAAs for AI inference: hhs.gov/hipaa/for-prof…

@MaziyarPanahi On-device PII detection with MLX on Apple Silicon is a huge deal for healthcare — HIPAA-sensitive workloads that can't touch the cloud now have a real path. 200+ models across 8 languages in v1.0 is serious coverage right out of the gate.

Medical AI models now run on iPhone. No cloud. No API. OpenMed 1.0.0 just shipped. MLX backend for Apple Silicon. Swift package for macOS and iOS. 200+ PII detection models across 8 languages. pip install openmed Open source. Apache 2.0.

@newlinedotco @MaziyarPanahi HIPAA compliance doesn't need on-device NLP and basically no one deploying AI in a regulated environment is using this strategy. All major model providers will sign a HIPAA BAA and agree not to train on patient data so no one is being blocked by this nowadays

on-device clinical nlp is the massive unlock for hipaa compliance. running pii extraction locally with mlx on apple silicon means phi never even touches a network interface, which basically deletes 90% of the security review friction for new medical apps. the really interesting part is the swift integrationprototyping a de-identification logic in a notebook and then dropping that exact same mlx artifact into a native ios app with openmedkit is a huge workflow win. most healthcare teams are still struggling just to get their cloud egress rules right while this setup lets you ship a private-by-default mobile emr tool in a weekend.

@takeiteqsy @MaziyarPanahi True but to be fair the example given is PII detection. The list of identifiers for HIPAA is pretty short (a good number are regex-able) and it doesn't take a gigantic model to detect a name or street address. HIPAA identifier list: dhcs.ca.gov/dataandstats/d…

@MaziyarPanahi A medical AI that runs on your phone is gonna be <=8b, I wouldn’t trust a model that small writing a throwaway script let alone my health. You’re absolutely mad if you download this.

@tannerlinsley @isninkhamiss @deckard_the_dev @tan_stack I’d love to help test them out and provide feedback

@isninkhamiss @deckard_the_dev @tan_stack They're private right now. We're working on them.

Vercel dropped agent skills for React best practices and it changed how I build So I made one for the @tan_stack ecosystem Best practices for TanStack Query, Router, and Start - packaged as agent skills for Claude Code and Cursor It's not official, I just built it because I needed it. But it's been incredibly useful Grab it below 👇