Inspired by Will Crichton's Good Ideas for Free and YCombinator's Requests for Startups.
-
CI verification platform for AI-powered robot simulations. Think 1 part Silogy, 1 part OpenAI Gymnasium, 1 part Elodin.
-
Publicly hosted Slack-clone where people use it to follow microblogs. Similar to Gitter.
-
Computer Science Metanotation as a programming language. Kinda already done by Shen.
-
Autocreated servers from an uploaded/drag-and-dropped Jupyter notebook with
notebook-http
mode.
-
It's an IDL language (like Protocol Buffers) that has:
-
Something like the
asymmetric safe schema evolution extensions from typical. I think they may have screwed up its implementation for choice - it should have prevented the production of a choice member rather than requiring a fallback.
-
Something like the inbuilt type-checker and refinement types from CUE
-
Something like the language-agnostic data validation runtime of CEL (with schema substitutability checked by the type-checker)
-
Some binary encoding format (like the Protocol Buffers encoding format or Flat Buffers) AND JSON encoding format that share similar semantics, so that you can use the same schema for both binary and JSON encoding.
-
No deserialization defaults.
-
Support for closed unions (abstract base class with a sealed set of subclasses), and open unions (TypeScript style). You'd be able to move fields between the abstract base class and the subclasses in closed unions, but not in open unions.
-
Enforced "semantic IDs" - e.g. fields with different names have to have different IDs in a closed union context.
-
Some stuff from tRPC. I dunno how that's going to work, but they've really nailed RPCs for TypeScript. It would be cool to have the rich tRPC-style schema modelling across any language.