Show HN: GlycemicGPT – Open-source AI-powered diabetes management | Hacker News Reader

Show HN: GlycemicGPT – Open-source AI-powered diabetes management

github.com

64 pointsjlengelbrecht61 comments

I'm a Type 1 diabetic and software engineer. Last year I went months between endocrinologists with no clinician reviewing my data. I'm an engineer, so I built the tool I needed — and now I'm open sourcing it. GlycemicGPT is a self-hosted platform that connects continuous glucose monitors, insulin pumps, and existing Nightscout instances to an AI analysis layer running on your own infrastructure. Data sources:

Dexcom G7 (cloud API) Tandem t:slim X2 and Mobi pumps (direct BLE) Nightscout (point it at your existing instance and you're running in minutes)

What the AI layer does:

Daily briefs summarizing overnight and 24-hour patterns Meal response analysis Conversational chat with RAG-backed clinical knowledge Predictive alerting with configurable thresholds and caregiver escalation

Important: this is monitoring and analysis only. GlycemicGPT does not deliver insulin, does not control your pump, and is not a closed-loop system. It reads your data and gives you insight on top of it. Your clinical decisions stay between you and your care team. Architecture:

Self-hosted via Docker or K8S — the GlycemicGPT stack runs entirely on your hardware BYOAI — bring your own AI provider. Use Ollama for fully local operation (no data leaves your hardware), or point it at Claude, OpenAI, or any OpenAI-compatible endpoint if you prefer a hosted model. Data flows directly from your instance to the provider you choose; nothing is routed through any centralized service operated by the project. GPL-3.0, no subscriptions, no vendor lock-in

Stack:

Backend API: FastAPI, Python 3.12, PostgreSQL 16, Redis 7 Web Dashboard: Next.js 15, React 19, Tailwind CSS, shadcn/ui AI Sidecar: TypeScript, Express, multi-provider proxy Android App: Kotlin, Jetpack Compose, BLE Wear OS: Kotlin, Wear Compose, Watch Face Push API Plugin SDK: Kotlin interfaces, capability-based, sandboxed

Looking for contributors — especially folks with BLE/Android experience or anyone in the diabetes tech space. Plugin SDK is documented if you want to add support for new devices. GitHub: https://github.com/GlycemicGPT/GlycemicGPT

Open articleView on HN

I'm a T1D who has an insulin pump looping with AndroidAPS and NightScout, what does this give you that Nightscout and Autotune doesn't give you?

And how do you deal with AI hallucinations?

I think the only thing that could be made better is tuning the I:C/ISF/Basal values automatically. And ISF is already handled by DynamicISF, while not perfect it reduces the variables you have to tweak.

Otherwise, when tuned correctly, oref1 et.al. provide amazing results and are safe. Hard to understand where I would use LLMs in this.

You sort of have that - not automatically though, but you can run autotune against nightscout and get a report of where things need to be adjusted. I run oref1 with DyanmicISF, and just run autotune every few months just to tweak values.

I genuinely don't see where I would use an LLM in this process.

I agree. For us, who are really good with software and all that, AndroidAPS is really good when you can easily learn how to use it.

Not that great for non-technical people. But it saved my life at least.

Edit: thanks for reminding me about Autotune. I used the azure app ages ago, but it's now integrated in AndroidAPS behind a secret setting.

If your already using AAPS and are happy with it then my honest answer is you don't really need this unless you want to get the data analysis layer. What you would get is AI analysis of your data, daily briefs, alerting, etc. I have written an honest comparison to other OSS tools today. This should provide some clarity on what this is and isn't. Let me know if you have any suggestions or ideas on what would make what I built more useful to you as am AAPS user. Happy to hear how the project can adapt and grow and fit into your usecase. - https://glycemicgpt.org/docs/platform/concepts/relationship-...

This is THE ONE domain where you would want to use classical machine learning and not unreliable LLMs. Unless you want to kill yourself, that is.

Yes, language has nothing to do with it and is complete overkill.

Probably something like SVM for warnings.

Unless the whole purpose is just daily reports.

I'm a T1D and tbh it's not that hard to manage, I just wouldn't need that. But for kids or the elderly, I see a use case.

The hardest to learn was that an unhealthy lifestyle resulted in a diabetes that was harder to manage. Too much carbs, not enough exercise, etc. After adjusting my lifestyle, it became quite easy.

The most pain, in my experience, comes from the discrepancy between the CGM - measured value and the prick-test value, even when accounting for time lag. I've used several CGMs and they've all been wildly off sometimes. I have a few T1D acquaintances who relied on their CGM alone and have significantly improved their HbA1c after accounting for that.

Maybe that information is useful to you.

Look into Eversense 355 (implantable), it has so much better accuracy for my wife than the Libre and Dexcom CGMs she has tried.

> But for kids or the elderly, I see a use case.

These are exactly the kinds of people who should NOT base their diabetes management on the "suggestions" of LLMs! There is the real risk that such users lack the competence or judgement to critically assess the convincing-sounding output of an LLM.

I don't know, Dexcom G7 is accurate enough. I do random tests from finger and it's about there a bit off but close. The previous versions were wildly off if you're having a hyperglycemia, and also G7 doesn't peak as high as measured from a finger. But in that level you already have ketones in your blood, and insulin resistance, and a bit extra insulin from the pump will not drop you to hypo.

Dexcom G7 has had inaccuracies for me. A standardized product meets non-standardized living beings, I guess.

Yeah. That's why my doctor lets her patients to test with different CGMs. They work differently with different people. For me Dexcom has always been best, but my doctor uses Libre because that suits her better.

This is quite possibly a horrible idea. Personal anecdote: ChatGPT once read a blood work report value as 40, when the actual report said 4.

So, I'm in the medical field building an EMR and LLMs have obviously been a really important topic in the industry the last few years. We're still not even sure that giving LLM-assisted suggestions TO ACTUAL DOCTORS AND CLINICIANS will be helpful let alone to the patient themselves.

It's breaking the golden rule of these tools which is to have someone with enough knowledge to verify the accuracy of the data it spits out. Patient's famously don't. Hell, even the actual staff don't really understand or know how these tools work (or the ways in which you can/can't trust them).

I don't think that LLMs are trustworthy companions in managing a complex metabolic disease like diabetes - especially if you deviate (ever so slightly) from the norm (very lean, very active, strict diet, etc.)!

I'm a T1D myself and like to experiment with ChatGPT (or Opus). My experiences are mixed

LLMs are overly cautious when it comes to correcting with insulin. They regularly advise against correcting before going to bed, even if this means that my blood glucose remains above 140 mg/dl for the whole night.

I am following a low to medium carb diet (<100g a day). ChatGPT always nudges me to consume more carbohydrates, even though I have a TIR of 90% (70-150 mg/dl). Why would I change my diet if it currently works very well for me? Still, most LLMs seem to favor carbs for some reason.

I am using Fiasp as my fast acting insulin. Typically, I inject around 1 to 4 IUs of Fiasp. Its glucose-lowering effect typically lasts for roughly 2-3 hours. Therefore, I know that it is safe to re-inject after three hours without risking insulin stacking. But ChatGPT regularly advises against that and wants me to wait another 1-2 hours.

I am not against automating diabetes management. In fact, I really appreciate projects that help with that. But I don't consider LLMs to be helpful in this regard. Their combination of training data bias, liability aversion, lack of context, and one-size-fits-all thinking disqualifies them from such tasks.

I understand this instinct, but I can see the appeal of capabilities that are well within the limits of a well-designed agentic system.

Imagine asking such a system, "look at my postprandial response to dosing for the past week and make ratio suggestions for breakfast, lunch, and dinner." This is genuinely helpful, saves time, and well within the reasoning limits. You could spot check if you like.

Is it worth setting up such an assistant for the value you'd get out of it? I guess that's on the user and how many similar use cases exist.

> look at my postprandial response to dosing for the past week and make ratio suggestions for breakfast, lunch, and dinner

I'm not so sure about that. A patient absolutely must critically evaluate the LLM's suggestions. A naive user risks severe complications. A user with that kind of competence, however, doesn't need an LLM for such trivial adjustments - they're obvious

[flagged]

(I'm a mod here) It looks like a useful project and I'm glad you submitted it, but please don't post AI-generated or AI-edited comments to HN - see https://news.ycombinator.com/newsguidelines.html#generated and https://news.ycombinator.com/item?id=47340079

Really appreciate this feedback. Every failure you named is real. I have had similar experiences with ChatGPT giving me bad information. I plan on solving for this by expanding on the RAG system that's being used. The AI gets its context based on settings you are giving it. The platform has some settings built in around insulin types but this needs to be expanded. The platform is not designed to give you dosing recommendations I really built this to help you understand your own data so you don't fly blind. Any thing the AI says to you is clearly labeled as not medical advice and the expectation is that your medical provider is in the loop before you make any medical decisions. I will take your feedback into consideration as I continue to explore new ways to shape the output you get from the platform.

Does it prompt logging? For example, when I was trying to monitor my BG after diagnosis, I tried to log my meals to correlate later, but 1) would forget and 2) wouldn’t have the energy to time align the stats. So a tool that even saw changes in BG and shot me a text or message (did you eat/exercise do something @ [time]?) and used the LLM or something else to capture and enrich the metadata. Paired with boring things like med reminders (I just realized I forgot my metformin while typing this) and giving me an easy visualizer with these meta points would be useful. If I’m tracking sleep on a device etc.

As others have said, the analysis might be risky. I don’t want to trust interpretation to anyone but myself (bear my own risk) or my clinician. But just remembering to capture the data and making it easily time alignable and possible augmentable in the future would be useful.

Yes this is actually on the roadmap. The project does not do this today as its still in alapha but what you are describing is exactly what i am building toward. Specifically behavior analysis where the AI notices changes in your blood glucose and asks you follow up questions to help you understand what you might have missed. I have had early adopters of the project specifically ask for this.

The alerts system and sharing with caregivers is a solved problem already (e.g. Dexcom's Follow, Abbot's LibreLinkUp).

Do you find the analytics actually helps? I.e. a lot of this will depend on what you ate and whether or not you logged it?

As a urologist who built and runs his own clinic management software, I'd encourage thinking about this question early: what does the system do when the LLM refuses to answer, returns malformed JSON, or hallucinates a glycemic value? In medical contexts, a 'silent failure' (system continues despite bad data) is much worse than a noisy failure (system stops and asks the user). The 'happy path' for an LLM-powered medical tool is usually well-designed. The failure paths are where the project lives or dies. Curious how you handle that.

Thank you. Really appreciate this feedback. I am actually already thinking about this. Hallucinations need to be clearly called out which is important. This is on the project's roadmap for me to address. There needs to be a way for users to clearly say "This is wrong. you need to reevaluate" In terms of alerting this is where we drift from what the LLM does and what the platform does. The platform already ingests data and stores it in the RAG system which provides the AI context but AI is the component that used for chatting about your data and providing you daily briefs. Alerting lives on the platform side so the AI may use it when we start implementing pattern detection to alert diabetics and care takers with questions such as "I see your glucose is rising. Did you eat and not bolus?" but for actual glucose events that fire during hyper or hypo events this is hard coded in the platform itself.

"This will all end in tears, I just know it"

Marvin

You know that current AI systems are not reliable and produce errors?

How do you protect your life and the life of others using your software against potential lethal errors?

Really nice of you to share this, well done!

About the risks, managing type 1 diabetes is exhausting, and most people will still sanitycheck the output alongside the hundreds of treatment decisions they make every day. That doesn’t change the fact that tools like this can nudge you to notice and look into patterns or things that needs attention.

Tools like this can also display false low glucose figures, leading you to reduce your slow-acting insulin and skipping fast-acting insulin. A day later, you start feeling nauseous (ketoacidosis), and you’re in danger of death.

I’ll keep using the manual glucose meters (like you advised), and would personally stay the fuck away from any transformer-based LLM to report medical data.

Looks interesting, being a Whoop user for the last few years, I have seen for myself that their AI Coach/AI based suggestions are a hit or miss 3 out of 10 times, slightly concerned about how accurate this will. Not a diabetic patient, but I do monitor my levels with a CGM from time to time, will definitely check it out!

The issue with Whoop’s AI is that there isn’t much data, and the data doesn’t have much prescriptive power, so it can’t really suggest anything useful. Recovery and Strain scores are made up, and even resting heart rate doesn’t tell you anything prescriptive for the day.

The data available to the LLM in OP’s app is the polar opposite. It’s all actíonable and real, so I bet it can draw more useful insights than Whoop reminding you that you didn’t exercise all week.

What’s the limit on badges in a README

Interesting. I can see the utility if you're going to see a nurse practitioner. But if your physician doesn't pull the actual charts for your device and visually inspect them.... try finding someone else.

Went through pregnancy with the mother having recently-diagnosed T1 diabetes – just barely not killed by grave neglect on behalf of healthcare due to how badly they missed the diagnosis to begin with.

On your work:

this is legit

it is appreciated

Hats off, I salute this, thank you

I mean this in the nicest way possible.

But if someone dies because this thing hallucinates their reporting - would you feel any sense of culpability?

“GPL says no warranty”

“People need to double check LLM output”

“You’re holding it wrong”

I really don’t know if we, collectively as a civilization, should be willing to accept this kind of hand-waving when it comes to creating things like this. Sure, tools make mistakes or people misinterpret reports without the help of LLMs - but LLMs are just on a whole other level where the mistakes are just part of how these things work from a fundamental level.

I don’t even trust AI scribes at my doctors office to transcribe my appointment due to errors. There is no way in hell I would ever use something like this that could just straight up lie about something that kills me if I get it wrong.

I've done this with the Libre 2 sensor. I added Gemini to it. It gets like 2 weeks of readings at once, and the user can "chat to their data". I added a meals tool as well, where the user can photo their meal, and the ai estimates the impact on the readings.

It's so helpful to offload some the thinking about the condition to ai, all these people moaning about 'muh safety' don't get it. T1D suffers have to think about it all day all the time. A person doesn't have their own blood glucose data in their head.

I truly appreciate your work (and I’ll absolutely take a look at it).

But let me say one thing: I’ve been diabetic for more than 20 years. Ten years of management with finger pricks, three measurements a day, and insulin pens (thinking about it now, it feels completely insane that anyone could imagine managing this madness in such a primitive way). Then came years of CGM systems (I’m on my third one now, with different types of sensors, but that’s not the point).

I tinkered, automated, hacked things. But in the end I came to one conclusion: you need a competent specialist (someone who also understands that we tend to be a bit tech-obsessed) who, besides listening to you, actually imposes a strategy. We are the ones who need to adapt to the mainstream approach so we can speak the same language and have methods that are compatible with “everyone else.” No doctor will ever fully understand your custom system, and meanwhile the key to proper management is not in what you built.

Precise carb counting (without cheating yourself), correct boluses given at the right times, marking exercise, boredom and repetition, and being lucid about the effects of changes (agreed upon beforehand!) to CGM settings — changes that should only be made when you’re certain you’ve been a “good and precise patient.”

I’m saying this from the perspective of my own devastated situation. I now have an HbA1c of 5.8, but only after 20 years of smashing my head against the wall (and suffering incredible damage, many mistakes, and the classic “I’ve figured it all out myself” approach).

Stay strong.

Life imitates comedy...

I'm just happy to see a GPL project.

The risk to benefits ratio of introducing a language model to interpret so clear signals is nowhere near justified.

Monitoring and analytics is important, but it is a solved problem. A language model will only be able to hallucinate about the relationship between meals and glycemic response. At best it does no harm, at worst it can directly misinform.

Yep. The oref1 algorithm is amazing and proven to make diabetic's quality of life better, AND SAFE. I don't understand why would you need to add AI to that mix.

But I will check this algo out. Maybe it has some interesting bits.

My experience is completely the opposite, of using LLMs to pattern match and cast diagnostic nets.

Is your perspective based on, say, opinionated principle?, or experience?

The benefits are enormous.

The risks; What risks? No diabetic with baseline adult competence is going to drive their insulin-delivery vehicle off a cliff because some app said so.

I think you're being too optimistic about your fellow humans' judgement. "Death by GPS" is a quite common occurrence: https://www.sciencedirect.com/science/article/abs/pii/S13550...

Type 1 diabetes with the sensors and pump technology that this software being presented here fits to is not Everyman Joe stuff. Someone who can set this up and get this going is already burdened with the kind of analysis that the app can assist with.

Risks:

Changing parameters on the insulin pump because the LLM said so

Neglecting to seek actual medical advice believing a LLM replaces it

Misunderstanding medical complexity (ie a prescription due to medical history not available to the LLM)

Are those different from the risks of reading about diabetes on Reddit or using Google?

> No diabetic with baseline adult competence is going to drive their insulin-delivery vehicle off a cliff because some app said so.

if you can't trust this thing then what is it doing? the implication that people that trust this software do not have adult competency is also confusing.

> Is your perspective based on, say, opinionated principle?, or experience?

your perspective is solely based on recent trauma so I don't know if it is more reliable in any capacity

Don't trust the thing. That's not what it's for.

Don't do as I say. I'm just a rando from the Internet.

Don't do as the author of the posted software does. Don't do what the software tells you either. But the software can certainly build an informative perspective and suggest patterns and movements in an exquisitely complex disease. Managing T1D with a pump is exhausting.

Second, re. "your perspective is solely based on recent trauma so I don't know if it is more reliable in any capacity"

This kind of statement is far beyond anything bounded by the self-respect of a balanced adult. What the fuck, and who are you?

My ex-fiancée almost died in 2020. We lost an unborn child in IVF due to grave neglect on behalf of healthcare who missed the glaringly obvious Type 1 diabetes she had; They never once checked her blood sugar. You know what I did? I read the literature. I read medicine, I read molecular biology, I read neuroimmunobiology, I read about the placenta and fetal development.

I stood by my fiancée and carried her by hand back to health. She recovered faster than the endocrinologists expected. Her pregnancy was exemplary, fullly intact placental vitals out to 38.5 weeks. Healthcare is in such a bad state that I was forced to interject and argue coolly and adamantly with doctors on several occasions about potentially severe mistakes they were about to make. EVERY SINGLE TIME when I interceded, it was confirmed correct by a second opinion from a senior doctor.

I don't come here speaking from trauma. I come here speaking from grim and serious and confirmed lived experience of stepping in and caring, without any margin for error. Know how you do that? With extreme humility and the utmost care.

Who are you to speak to me like that; I can tell that you know not at all who I am or what I have been tasked with in this life, because then you would not. talk. to me. this way. Okay?

> The risks; What risks? No diabetic with baseline adult competence is going to drive their insulin-delivery vehicle off a cliff because some app said so.

My local physician says otherwise, with respect to facebook posts about dosages. I'm convinced the same applies to LLM generated content with respect to people blindly following the computer.

I ask for your understanding in that I chose to have "baseline adult competence" highly load-bearing in my comment. This does not include people who have only such poor judgement to guide them that they use Facebook posts as input to managing their T1D pump/sensor-based management.

It is entirely possible to beneficially and safely use software like the that which is the topic of the post.

> No diabetic with baseline adult competence is going to drive their insulin-delivery vehicle off a cliff because some app said so.

You 1000% don't work with the general public in a tech way.

"Baseline adult competence" was load-bearing there.

This is not an app for the general public.

> This is not an app for the general public.

Can you point me to where it explicitly says that this app is not to be consumed by the general public? Or explain how that could even be enforced?

And my entire point was "baseline adult competence" means very little. Competent in what? Technology? Insulin administration? Both? If they're competent in technology but not insulin administration, than this is obviously a bad idea. If they're competent in insulin administration, but not technology, then why would they use it?

We're not even at the point where we can definitely say it's a good idea to surface this information to actual professionals let alone someone with no clinical experience.

It's a bad idea, period. I work with both clinicians and the general public and the idea that this can be responsibility used by either is a pipe dream that only people who work with neither can believe in.

This is as dangerous as people talking together on e.g. Facebook groups. I've seen people give flat out dangerous advice, and other people take it.

Nursed someone back to health and went through a T1D pregnancy and certainly had to work with the clinicians during; Desperately: There is use for this software, and no greater need to bury it or unmake it than it is to bury or unmake Facebook groups or other discussion venues.

In fact, software that actually reads literature and presents perspectives actually derived from plausibility from a scientific basis is in many cases a 100% increase of availability of such a perspective. Doctors – from experience – may not offer that. And forum lore rarely does.

Software that actually reaches into literature – which LLMs do – and presents its perspective with the required framing is arguably a strict improvement to the infohazardous environment of a given patient with a difficult condition like T1 diabetes.

And I ask for this discussion's participants' understanding of the sarcastically wry use of "basic adult competence". It was at once arrogant and caring, and put forward… in something as quasi-serious as a HN discussion. I think that's OK.

> his is as dangerous as people talking together on e.g. Facebook groups. I've seen people give flat out dangerous advice, and other people take it.

Not even remotely close. A large portion of people in the field still don't understand how LLM's work let alone your grandmother on Facebook. They're apples and oranges.

> There is use for this software, and no greater need to bury it or unmake it than it is to bury or unmake Facebook groups or other discussion venues.

Discussion forums are nothing like an LLM. People don't assume the other person on the forum is magic. Again, these are totally different things that have no correlation with each other. People have been in some form of forum for hundreds of thousands of years. They have no experienced a genie that has ingested the entirety of human knowledge. They're just flat out 2 different things.

And you're experience as a single person doesn't mean that experience is universal. I quite literally work with clinicians and patients every day. Clinicians aren't evne sure how to use these tools (with actual clinical knowledge), but yet we're ready to shoot them out to consumers? It's wild.

Of course there is a need to get better info, but that doesn't then follow that the route of that should be an unregulated, unlicensed LLM.

Thanks for calling out!

We're even yet debating and trying to understand what impact AI has on software engineering and quality let alone putting AI into something that's directly linked to a human's well being.

That's just risk/benefit to the user. As the developer, I'd be concerned that publicly distributing and marketing this, even with a GPL "no warranty" license and even free to the user, is illegal.

You can distribute the source code for research purposes legally at least here in Germany. That's how AndroidAPS is even possible.

FDA approved?

No but the project has undergone an in depth evaluation from Open Collective prior to being accepted for fiscal hosting. Additionally I am in contact with an Attorney who is reviewing the projects stance. The project as it is right now is aims to not cross any boundaries that would make it a medical device. The platform does not control insulin or make dosing recommendations. It's still in early alpha but I have made my position clear. The platform serves as a tool to help you understand your data and help you have better discussions with your medical provider.

Reply on news.ycombinator.com