3 ms·

We currently support AWS but use terraform for deployment, which is pretty cloud agnostic. So far, we haven't gotten any major requests for expanding to other c

We currently support AWS but use terraform for deployment, which is pretty cloud agnostic. So far, we haven't gotten any major requests for expanding to other cloud providers and most of our incoming customers are already on AWS anyways. One of the main reasons why healthcare players were moving onto Azure was for in-built HIPAA compliant OpenAI access. We've been able to help our customers directly sign BAAs with OpenAI so this wasn't a concern.

DaiPlusPlus 3y ago
> for in-built HIPAA compliant OpenAI access Sorry but what?
- selinkocalar 3y ago
  A lot of healthcare companies are wanting to use GPT4, Whisper, and other LLMs from OpenAI, but these aren't HIPAA compliant out of the box. You still have to sign a BAA with OpenAI and get on their zero data retention (ZDR) plan. Because of the close partnership with Microsoft and OpenAI, Azure makes it easy to get HIPAA compliant access to certain OpenAI models without having to go through OpenAI directly. This is why a lot of AI healthcare companies were building on Azure at first. Hope this clarifies!
leetrout 3y ago
> We currently support AWS but use terraform for deployment, which is pretty cloud agnostic. Nothing could be further from the truth.
- doctorpangloss 3y ago
  You’re right but the kind of people who buy this service buy into the perception that Terraform is cloud agnostic, and perception is reality. Another POV is that the compliance companies sell the holistic social experience of compliance. The people for whom this matters need checkboxes and don’t mind paying for consulting disguised as a CSA. In fact they may even prefer it.
  - selinkocalar 3y ago
    At the end of the day, whether its AWS S3, GCP Buckets, Azure Blob Storage, or EKS, AKS, GKE — there are services to serve as drop in replacements for their competitors. The principles of deployment will remain the same as it pertains to access and compute allocation, but obviously the resource definitions, and the definition syntax, is unique. We’re solely focused on building out our AWS support (a function of market pull), and plan to use the same architecture design for additional cloud providers when it comes time to make that jump. Completely get where you’re coming from w.r.t. organizations looking to check a box. If we wanted to sell to those organizations, we would stop short of architecting their infrastructure and implementing best practices. A checkbox is never enough, and a point in time review can be easily gamed.
deathanatos 3y ago
While you're right about the access to AI models on Azure, I wouldn't tie my infrastructure to Azure just for that. Sure, you might have to use them for that service, but ship the queries in & the results back to a cloud of choice; especially LLMs in most uses cases won't represent a huge amount of data; the costs of egress < the costs of dealing with Azure. We're (now only partially) on Azure for reasons stated upthread: desire from the industry. We've partially moved since, and the experience as a whole has soured me on Azure as a reliable cloud provider (I've had to engage with support far more often). Their support, in particular, is terrible¹, and a lot of their offerings IME are less stable or harder to work with than AWS. I'm now moderately experienced with GCP, and I think GCP has them beat, too. ¹missed SLAs, a huge desire to close tickets prematurely, failures to address things asked of them, really bad communication skills, constantly divorcing replies into new threads, dredging up asked & answered stuff, repeatedly asking questions covered in the original ticket's opening set of answers, asking the customer for stuff they should know, etc.
- selinkocalar 3y ago
  Sorry to hear that your experience with Azure has been sub-par. I know there was a sizable wave of people that migrated over to Azure for the AI models but haven't heard too much from them since. Interesting to hear about your experience...

DaiPlusPlus 3y ago

> for in-built HIPAA compliant OpenAI access Sorry but what?

selinkocalar 3y ago
A lot of healthcare companies are wanting to use GPT4, Whisper, and other LLMs from OpenAI, but these aren't HIPAA compliant out of the box. You still have to sign a BAA with OpenAI and get on their zero data retention (ZDR) plan. Because of the close partnership with Microsoft and OpenAI, Azure makes it easy to get HIPAA compliant access to certain OpenAI models without having to go through OpenAI directly. This is why a lot of AI healthcare companies were building on Azure at first. Hope this clarifies!

leetrout 3y ago

> We currently support AWS but use terraform for deployment, which is pretty cloud agnostic. Nothing could be further from the truth.

doctorpangloss 3y ago
You’re right but the kind of people who buy this service buy into the perception that Terraform is cloud agnostic, and perception is reality. Another POV is that the compliance companies sell the holistic social experience of compliance. The people for whom this matters need checkboxes and don’t mind paying for consulting disguised as a CSA. In fact they may even prefer it.
- selinkocalar 3y ago
  At the end of the day, whether its AWS S3, GCP Buckets, Azure Blob Storage, or EKS, AKS, GKE — there are services to serve as drop in replacements for their competitors. The principles of deployment will remain the same as it pertains to access and compute allocation, but obviously the resource definitions, and the definition syntax, is unique. We’re solely focused on building out our AWS support (a function of market pull), and plan to use the same architecture design for additional cloud providers when it comes time to make that jump. Completely get where you’re coming from w.r.t. organizations looking to check a box. If we wanted to sell to those organizations, we would stop short of architecting their infrastructure and implementing best practices. A checkbox is never enough, and a point in time review can be easily gamed.

deathanatos 3y ago

While you're right about the access to AI models on Azure, I wouldn't tie my infrastructure to Azure just for that. Sure, you might have to use them for that service, but ship the queries in & the results back to a cloud of choice; especially LLMs in most uses cases won't represent a huge amount of data; the costs of egress < the costs of dealing with Azure. We're (now only partially) on Azure for reasons stated upthread: desire from the industry. We've partially moved since, and the experience as a whole has soured me on Azure as a reliable cloud provider (I've had to engage with support far more often). Their support, in particular, is terrible¹, and a lot of their offerings IME are less stable or harder to work with than AWS. I'm now moderately experienced with GCP, and I think GCP has them beat, too. ¹missed SLAs, a huge desire to close tickets prematurely, failures to address things asked of them, really bad communication skills, constantly divorcing replies into new threads, dredging up asked & answered stuff, repeatedly asking questions covered in the original ticket's opening set of answers, asking the customer for stuff they should know, etc.

selinkocalar 3y ago
Sorry to hear that your experience with Azure has been sub-par. I know there was a sizable wave of people that migrated over to Azure for the AI models but haven't heard too much from them since. Interesting to hear about your experience...