Imagine a Weir Studio

This week, I re­ceived a mes­sage from some­one work­ing on a learn­ing man­age­ment sys­tem. Ap­par­ently, they use harper.js to do spell and gram­mar check­ing on in­ter­nal doc­u­ments. They wanted to be able to con­fig­ure Harper to catch a spe­cific class of mis­takes that are com­mon to their work­flow. Look­ing through the doc­u­men­ta­tion for harper.js, they could­n’t see a way to do what they wanted.

Naturally, I di­rected them to­ward my re­cent work on Weir, an ex­pres­sive pro­gram­ming lan­guage for lo­cat­ing er­rors in nat­ural lan­guage. For those just catch­ing up, this post is a part of a se­ries. You should­n’t need to go back and read my pre­vi­ous posts, but I’ll link them here any­how.

• Building The Weir Language
• Updates on the Weir Language
• Generating Weir Code With LLMs
• Imagine A Weir Marketplace

Unfortunately, I had to in­form them that Weir is still an ex­per­i­ment (although it’s in the master branch and thus used by nightly users) and there­fore is­n’t yet avail­able through pub­lic chan­nels. That said, it’s an ex­per­i­ment I’d like to make avail­able to the pub­lic soon, along­side the Weir Marketplace. Before I do so, I want to make it as easy as pos­si­ble to au­thor rules us­ing Weir. That’s why I’m in­tro­duc­ing the Weir Studio.

The Weir Studio

The Weir Studio comes from a pretty sim­ple idea. We want to give peo­ple all the tool­ing nec­es­sary to build and dis­trib­ute Weir rules with­out any up­front cost. That means there should be no tools that need to be in­stalled, no spe­cific op­er­at­ing sys­tem set up, and no pro­gram­ming ex­pe­ri­ence needed.

The so­lu­tions to the first and sec­ond items are sim­ple: Just like any other com­po­nent of Harper, we can com­pile Weir to WebAssembly and con­nect it up to a web ap­pli­ca­tion. Us­ing off-the-shelf com­po­nents for ter­mi­nal em­u­la­tion, text edit­ing, and vir­tual filesys­tems, we can pre­sent a mean­ing­fully fea­ture-rich de­vel­op­ment en­vi­ron­ment from in­side the user’s browser.

The last item is also straight­for­ward. I’ve demon­strated that LLMs are al­ready ter­ri­bly ef­fec­tive at writ­ing Weir rules. By mak­ing that sit­u­a­tion ob­vi­ous and ac­ces­si­ble, we can em­power peo­ple to write and dis­trib­ute their own Weirpacks (files con­tain­ing one or more Weir rules) with­out any prior un­der­stand­ing of the lan­guage or even of ba­sic pro­gram­ming prin­ci­ples.

The Weir Studio is still in progress, but the ini­tial ver­sions are al­ready prov­ing ex­tremely ef­fec­tive for some of my other work.

Putting Power Back into the Hands of Users

A com­mon frus­tra­tion with tools like Grammarly or QuillBot is that they lack cus­tomiza­tion, and thus are of­ten un­able to di­rectly ad­dress user’s needs. By al­low­ing users full ac­cess to the in­ter­nals via Weir, we can help them help them­selves while also en­cour­ag­ing peo­ple to con­tribute back to the pro­ject. It’s a win-win!

What’s Next?

I’d like to wire this up to a mar­ket­place as soon as pos­si­ble (see what I did there?) to stream­line the pub­lish­ing process. More crit­i­cally, I’ve been in­formed that the in­clu­sion of dic­tio­nary files di­rectly into Weirpacks should be pri­or­ity num­ber one. I’ll likely start tack­ling that as soon as I have a spare mo­ment.

If you have thoughts on where Harper or Weir is go­ing, please don’t hes­i­tate to voice them.

---

P.S. A great piece was just pub­lished about Harper.