Has anyone here tried using the AI Services in Microsoft Azure? I would appreciate any advice. By the way, I’m a newbie with weweb. The idea is to use a live camera, feed it to the model and in real time grade the object and count the object based on the grade level. TIA!
Oooh sounds like a cool project! Wish I had the skills to help 
Maybe @caffeinatedwes or @Zacharyb can? (Yes, I’ve pegged you two as my go-to AI/WeWeb guys 
)
2 Likes
Haha, thank you, @Joyce!
Our team has not used Microsoft Azure’s microservices. However, it seems like you’re looking for specific technical guidance to build this yourself. In that case, I highly recommend @raydeck’s StateChange community.
3 Likes
Hello !
So i will try to be synthetetic , first question important do you currently use any AI services from Azure currently whereas Document Intelligence, AI Search, CosmosDB (if yes is it the NoSQL, Postgres, Gremelin or Cassandra ?)
Do you use Custom Vision or another specific service or you intent to use it with a basic Azure OpenAI endpoints ?
But if you haven’t yet explored the services, know that is it not cheap at all, especially with the kind of your use case (realtime with live camera into a classification ML) would probably a way to mitigate the cost with using a custom ml.azure flow, but at the end know that i hope you have at least few k$ credits !
The thing i couldn’t really be sure, what kind of realtime tech to use, tensorflow should have interesting thing around it or prob directly azure. The part where you go from weweb/user request in live stream two thing come in mind :
A trained or pretrained Opensource model (actually pretty great decent progress with small model) it will need some cloud calculation
It really depend on how you count to use live camera first through WeWeb, and what is the usecase in terms of recognition, the prior datasets and evaluation if relevant, and then the whole continious teachable LLM app to make the recognition and classification more precise.
The most simple advise i would tell you is to go on the github of microsoft, and also the azure one and look for a template that could match your needs !
For the rest, yeah don’t hesitate if you have question !
( Qredence & Qream if you are interested, (there’s is almost no one but it’s free 100% and answer with the best of my knowledge, without going on areas where i wouldn’t have a answer
Anyways, hope the readiness of this message will be clear enough to understand !
2 Likes
some reessourcce
Can you tell more about the grading and counting challenge? I suspect the big challenge here is the video capture and moving those frames over the wire at some reasonable pace to fulfill your object detection/evaluation needs.
I think about a few things:
- Capturing video on browser is a bit dodgy and often something you’r trying to do with a mobile. It’s not usual web code. CHeck out how you feel that experience is with a browser. Possible you might find quickly that you want a native app for that. Either way it lets you “shift left” a key shipping consideration.
- Pre-processing or filtering the video locally is a serious option. This could significantly reduce the number of frames you’re sending to whatever analytic service you apply.
- There is no universe where this is going to be “beginner” level stuff. Which is fine! Bring to bear your maximum resources of chatgpt, web searching, communities like this one (and MSFT etc). And as @caffeinatedwes kindly pointed out, we work on these kinds of hard questions during office hours at State Change.
2 Likes
I think Tensorflow is a way, simple, can take live camera input and has some additionnal way of building it
an example in the link.
Beside that, for classification you can do it if it’s not complex, and you think you got enough data
anyway there is many way to do it (Create ML for some local inference for classification for eg)
in the author context, azure should be used not to build but to deploy his existing model
1 Like
And for those that want to just try simple introduction personnal project https://teachablemachine.withgoogle.com/ it simple and easy to use (not for production of course)
and we are in ML ofc
Thanks Joyce!
Hi! I’m quite familiar with Azure AI. I’m open to other AI services or options. Thanks for your response!
1 Like
Thanks, Raydeck! I’m looking at HD cameras and AI Inference systems probably using Nvidia Edge systems to process images or use a Cloud AI service. Azure came to my mind because of my Azure experience. Essentially, the challenge part for me is how to build the whole webapp (Poultry Management System - later we will also use AI to check the status of chicken) from capturing those images to processing and displaying the charts.
Hi @Zacharyb, thanks for sharing all this information. Would you mind giving some tips if I develop an app on Weweb and use TensorFlow or Teachable Machine? I would like to know how the interplay will look like for my case–real-time sending of video images and getting the data from ML service.
The challenge is the whole app? I recommend you take it in parts and get your feet wet. I’d start with playing with the most AI part on the back end: Azure cognitive/AI services , and the most “weird” part at the front end - capturing frames from your video input component on the front end. Then you can start to stitch these two towards each other with table stakes software like weweb.
1 Like
Teachable Machine is just for education to show example on how it’s work ML basicly
For your usecase, it really depend on your ability to train your model prior to deploy your model at first.
Once you got your database and the overall MLops define, you can use compute hours on Azure services to have a cloud ready model that would be “production ready”
other thing i note is the use of the camera which bring me to the question : Is weweb really the tool you need ?
I might better go for tool that have been designed for interacting with camera sensor etc natively.
I understand that I can get the whole stack of AI services + dashboard from one provider but from my experience with IIOT packaged solutions in the F&B manufacturing, the investment could be quite high. Just wanted to know if anyone out there has done the same solution using Weweb. Absolutely, the interface can be done by a low code dev platform. I just don’t have time to explore and develop quickly. Thanks for your suggestions, appreciate it much. Cheers!
Honestly i would try to explore solutions from Power Apps, Power automate, for eg, would make sense imo in your case
1 Like
Thanks for the suggestion. Yes, it would make sense for customers running on Microsoft 365. Considering cost, now I’m looking for an alternative solution for this use case.