@ljacobsson.bsky.social
AWS | Serverless | Founder of Kvist.ai | Independent consultant | Table tennis
Nice!
I've built Kvist.ai which is a polished version of this.
It's still on AppSync v1.
It's currently using OpenAI because Bedrock is still too buggy and not yet ready for production.
Working on an integration with a SaaS that gives me HTTP status 429 whatever I do.
The best way to find out why they don't scale is to look at their job openings.
#serverless
Whats's going on #AWS?
Can we recommend our clients to use Bedrock for their AI workloads moving forward?
Things break - even at AWS, but what's alarming and unprecedented by this is the complete lack of comms around it. We had to find out via our users who experienced a drop in quality.
Contacted the support who doesn't seem to prioritise getting us back online... ๐งต
On Oct 20, Amazon Bedrock had an issue that strangled access to the Bedrock models we prior to that had full access to. This happended to lots of customers.
We went from 20rpm to just 1rpm without prior notice and no comms after. We have a fallback on OpenAI, so our app carried on working... ๐งต
Well, can't say I've been on top of my tooling work in the past year, but this one triggered me to hack-attack
24.11.2024 09:43 โ ๐ 1 ๐ 0 ๐ฌ 0 ๐ 0I've just built a first stab at an instant converter between JSONPath and JSONata ASL.
Still some corner cases missing, but does most of the work.
lars.tools/jsonata
#aws #stepfunctions
...and it finally happened! Been waiting for this all my life!
aws.amazon.com/blogs/comput...
So many really good #preinvent announcements so far and they've been the opposite of GenAI focused.
This is the good old #AWS that got me hooked almost 10 years ago โค๏ธ
aws-news.com/custom_feed/...
This comes almost on the day one year after it was announced that Mathem was getting acquired and having its serverless infra swapped out for k8s.
Most of Mathem was .NET ฮป and it performed way better than the k8s replacement.
Great w/ snapstart, but make sure you actually need it before adding it
In case you're looking for serverless people on here, I made a starter pack for you.
go.bsky.app/PmjTiyJ
I swapped to N.Virgina and my invocation went through! Once - the rest got throttled.
My client wanted this inference done ASAP and gave me an OpenAI API key.
I will move it over to Bedrock once I get this sorted, but why, #AWS, does it have to be so difficult to start using Bedrock?
Today I worked on a client's account. I saw that Claude 3.5 Sonnet had a green Access Granted in both Frankfurt and N.Virgina. I thought "nice - someone has already done the support ping-pong" and started building. I started in Frankfurt, but every request got throttled.
13.11.2024 20:38 โ ๐ 0 ๐ 0 ๐ฌ 1 ๐ 0
Under account & billing support there's a Bedrock Allowlisting option. I'm going with that. A couple of days later I hear back from someone who said they passed the issue on to tech support.
Another day goes by and voila, I have access.
I wanted to enable Claude 3.5 Sonnet in Frankfurt because an official #AWS blog post said it's available there.
It wasn't available, so I had to contact support. It's a brand new account and I'm on basic support, so can't raise a technical support ticket.
I wonder how many #Bedrock users #AWS lose to OpenAI due to how tedious it is to get started with Bedrock on a fresh account and how easy it is to get started with OpenAI.
I like Bedrock once I'm given access, but this is not good user experience.
My experiences with Bedrock in ๐งต
Kvist.ai's #preInvent quiz #2 is live @ kvist.ai/preinvent
We'll publish a new quiz every Friday leading up to #AWSreInvent and then daily during the conference with some potential prizes.
Generated from @lucvandonkersgoed.com's brilliant aws-news.com with help from also brilliant #AmazonBedrock
