Show HN: Cardamom – Deploy ML to AWS Lambda with a Function Call (opens in new tab)

Hi HN!

I wanted to share my machine learning compiler, hosting ML models on AWS Lambda.

If you go to the linked page, there are instructions to run a script which generates a model using sklearn, feeds it to my endpoint, and then calls the created endpoint on lambda. In case you're unable to run a script, I've also included a video on the page.

Unlike other ML hosting services I've seen, where models are spun up behind containers, this service compiles the model down to either a static C library or WASM library, and then has a template lambda function do the HTTP parsing / argument handling / etc.

Upsides of this approach:

  - You can directly embed models in applications, callable via FFI
  - You can run ML in weird places, like mobile apps, embedded devices, or the browser

Downsides:

  - I have to implement the model inference for each algorithm
  - As such algorithm support is limited right now

My goal is to make the interface easy enough that anyone who can build a model can use it to deploy their model somewhere it can deliver value, rather than having to request an engineer's help.

I'm fine hosting models for now, but I'm also excited for building out custom integrations like running the models on mobile apps, inside existing applications via FFI, or even embedded devices.

Please let me know what you think!