Hugo Dutka

[Hugo Dutka @hugodutka]

TIL that companies on LinkedIn have numerical ids and LinkedIn's is 1337. https://t.co/R6uBjoS5dk

[Alvaro Molina @Alw3ys]

Announcing @DeployPlex, The native FastAPI platform.

In December 24, 2018 @tiangolo released @FastAPI, and two years after it's been downloaded more than 800,000 times, he didn't expect it to work, but it did, and now he has enabled many teams across different organizations to build APIS for Machine Learning and data applications.

Another massive adoption and traction @FastAPI has gained is from new developers building LLM applications leveraging @OpenAI APIs, vector databases and the emerging fast growing open-source framework @langchain.

These components combined has made extremely easy to get up an running on your computer and built quickly a LLM backend.

The new blocker
However, the challenges have evolved. While it's now possible to prototype an application with approximately 5 lines of LangChain code and another 5 lines of FastAPI to expose an endpoint, the journey from prototype to production remains fraught with obstacles. Developers are finding themselves retracing the same steps, and organizations are repeatedly building infrastructure on top of AWS, Google Cloud, and Azure. Not only are they hosting their applications on these platforms, but they're also investing significant time in automating CI/CD processes and self-hosting vector databases - a crucial component of LLM-based apps.

But this journey isn't without its toll. The investment in time and resources is substantial. Transitioning an application to production involves myriad tasks: managing database backups, intelligently caching builds to ensure build times don't hinder productivity, handling SSL certificates, managing container registries, and optimizing VM sizing. All these tasks are undertaken with the aim of maintaining clean Infrastructure as Code (IaC) practices, like using Terraform. And just when things seem to be running smoothly, unexpected changes, like a licensing shift, can throw a wrench in the works.

What we are not building
We are not monitoring debugging or monitoring your LLM prompts, we are not helping you test and evaluate your dataset curation, we are not helping you manage your prompts.
If you want to do that use LangSmith, from @langchain, that's their play, and hopefully we can forge some partnership in the upcoming future.

Breaking Down What We're Building
Our vision with @DeployPlex is to create a platform tailored for engineers who are passionate about building top-tier AI products at the application layer. We're staunch believers in the power of Python, and we're doubling down on it. We're leveraging @FastAPI prowess as a modern, high-performance web framework for building APIs.

With @DeployPlex, you can expect:
- GitOps philosophy.
- Zero-Config build, test and deployments and in less than a minute
- Seamless app rollbacks and autoscaling.
- Batteries included Vector DBs. Backups, autoscaling (more to be announced soon).
- Managed SSL Certificates.

Let's face it: managing infrastructure isn't a walk in the park when you want to serve thousands or million of users compared to local prototyping.

What's our vision?
To build the platform AI-native apps.

Our beta access
Hold onto your hats! We're not fully launching just yet. But rest assured, we're gearing up for it. In the meantime, we're eager to engage with our community. Request beta access on our site, and we'll be selecting 100 engineers to participate in our beta program.

We are exited to help you build faster AI-native apps.

[Hugo Dutka @hugodutka]

@fruit_case @PierreZ We did. Here's a GitHub issue explaining why we didn't use it: https://t.co/CukVlf9k1b