I am currently in the process of implementing a TLV message protocol (think MsgPack, CBOR, etc).

This is something I've been kicking tires on since my time at $BIGCORP; JSON without the bloat, Protobufs without the ceremony. I've drawn a lot of inspiration from MsgPack, CBOR, and Ion 1.1. Big emphasis on a tight set of core primitives, low-cost extensions, storing reused values/schemas, optional pre-negotiation, etc. That said, I've now been spending time trying to study the performance angle to make sure the design doesn't have a negative impact on encoding/decoding performance before committing to the implementation.

Regrettably nothing much to show (at least yet), but hopefully if nothing else it will become my go-to format for other personal projects that I work on.