anuragvishwa

Hey Guys, could you please help me choose between 2 ideas i am planning to build:

1.Prompt Engineering Evaluations for LLMs through playground .

2. Fine Tuning a foundational model.

Thanks.

These are the features i am working on - 1. Collaborative playground with history 2. Prompt management and deployment controls 3. Backtest changes and confidently update models 4. Capture feedback and run quantitative experiments

Would like your feedback and suggestions

Launching soon, the playground : 1. Collaborative playground with history 2. Prompt management and deployment controls 3. Backtest changes and confidently update models 4. Capture feedback and run quantitative experiments

Feedbacks and comments are most welcome

I am planning to build a startup around some pain points around K8s, it may be related to platform engineering as well. - I tried building Opensource Platform as a service (Visualith) that would help you deploy k8s cluster and application on your AWS account but have 0 takers. I think i am missing the point here. Need some feedback around the pain points so that i could build an actual solution.

I recently launched my open-source product, visualith.com . It is a PAAS for AWS. - It helps you quickly deploy cluster(EKS) and applications using GitOps principle (ArgoCD). - It contains graphana and prometheus pre configured. - It also contains Kubecost preconfigured. I am unable to get a single user to try the product. What am i doing wrong or is the timing wrong. Desperately need some suggestions.

With comparison to say Kubecost , the product will have - - Muticluster visibility in the single dashboard (for Kubecost , muticluster is enterprise ). - The application would also contain apart from monitoring Autoscaling and bin packing ( based on VPA recommendation for each application and providing a window to automatically adjust it). - Smart cost optimization : Tools like Karpenter(AWS) would be used to quickly get the alternatives to On-Demand instances and providing spot instances for them. - Better UX - The UX would be our top priority with 5 min, setup , maybe with a helm package. - Automated rightsizing : Based on the VPA recommendation . These are the basic points i could ponder on for the application. 1. The monitoring part would be open-source and free for multi clusters. 2. Report generation and alert management to (Slack,email etc) open-source. 2. The autoscaling part/Bin packaging/Node autoscaling/Spot instances would be enterprise features. Would love to know your insights and feedback . Thanks for going through the context

Sorry for the confusion, but by single command , i meant to say single Helm command!

With comparison to say Kubecost , the product will have - - Muticluster visibility in the single dashboard (for Kubecost , muticluster is enterprise ). - The application would also contain apart from monitoring Autoscaling and bin packing ( based on VPA recommendation for each application and providing a window to automatically adjust it). - Smart cost optimization : Tools like Karpenter(AWS) would be used to quickly get the alternatives to On-Demand instances and providing spot instances for them. - Better UX - The UX would be our top priority with 5 min, setup , maybe with a helm package. - Automated rightsizing : Based on the VPA recommendation .

These are the basic points i could ponder on for the application. 1. The monitoring part would be open-source and free for multi clusters. 2. Report generation and alert management to (Slack,email etc) open-source. 2. The autoscaling part/Bin packaging/Node autoscaling/Spot instances would be enterprise features.

Would love to know your insights and feedback . Thanks for going through the context.

It is build in progress, wanted to know the feedback/need of the product before launching. Nevertheless, it would get launched by Jan 30th 2024 .

The sample page would look like this - https://www.reddit.com/media?url=https%3A%2F%2Fpreview.redd....

I would keep posting the updates to the project here. Thanks for going through the context.