Hope everyone's doing great! So, I've been thinking a bit about our TypeScript client and how it's currently a mirror image of Weaviate's methods. It's cool and all, but it kinda feels like we're speaking a different language than our target users β the non-AI devs out there.
What's on My Mind
Our current method names are straight out of the AI textbook, which is great for the AI pros but maybe not so much for web developers and others. Take "nearText", for example. Makes total sense if you're into LLMs and NLP, but to a regular dev, it might as well be hieroglyphics.
Some Quick-fix Ideas
I was playing around with some ideas to make our methods more like everyday language. Here's what I came up with:
Let's call nearText something like search. Simple, right?
How about renaming hybrid to find?
For generate("groupresults", ...), what do you think about generate.fromMany?
And generate("singleresult", ...) could be generate.fromOne.
Lastly, similarImage could be way cooler as nearImage.
Making Things Easier to Use
Also, I was thinking, why not jazz up our methods a bit? Like, our nearImage (or similarImages, hopefully) could be way more flexible. Why limit ourselves to base64? Let's throw in support for image-url, buffer, `? and other formats that front-end folks love.
Lighter and Faster?
And just throwing it out there β what if we had a lighter version of our client? Something thatβs super easy and straightforward, focusing only on the most popular stuff. Like, changing unbody.get.googleDoc to just unbody.googleDoc.
Your Thoughts?
Iβd really love to hear what you guys think about this. Are these changes doable? Would they help? Got any tweaks or totally different ideas? Let's chat about it!
Hey Team π,
Hope everyone's doing great! So, I've been thinking a bit about our TypeScript client and how it's currently a mirror image of Weaviate's methods. It's cool and all, but it kinda feels like we're speaking a different language than our target users β the non-AI devs out there.
What's on My Mind
Our current method names are straight out of the AI textbook, which is great for the AI pros but maybe not so much for web developers and others. Take "nearText", for example. Makes total sense if you're into LLMs and NLP, but to a regular dev, it might as well be hieroglyphics.
Some Quick-fix Ideas
I was playing around with some ideas to make our methods more like everyday language. Here's what I came up with:
nearText
something likesearch
. Simple, right?hybrid
tofind
?generate("groupresults", ...)
, what do you think aboutgenerate.fromMany
?generate("singleresult", ...)
could begenerate.fromOne
.similarImage
could be way cooler asnearImage
.Making Things Easier to Use
Also, I was thinking, why not jazz up our methods a bit? Like, our
nearImage
(orsimilarImages
, hopefully) could be way more flexible. Why limit ourselves to base64? Let's throw in support forimage-url
,buffer
, `? and other formats that front-end folks love.Lighter and Faster?
And just throwing it out there β what if we had a lighter version of our client? Something thatβs super easy and straightforward, focusing only on the most popular stuff. Like, changing
unbody.get.googleDoc
to justunbody.googleDoc
.Your Thoughts?
Iβd really love to hear what you guys think about this. Are these changes doable? Would they help? Got any tweaks or totally different ideas? Let's chat about it!