I have a good feeling about this

Hey everyone! I’m a software engineer working with LLM agents, RAG systems, MCP integrations and data pipelines. I can help troubleshoot unreliable agent workflows, improve retrieval quality, investigate API or database bottlenecks and fix deployment issues. If you’re stuck on something or need another engineer on your project, feel free to DM me with what you’re building and where you need help.

was just watching this and I was thinking that there could be some value in having one agent define the parameters of a problem and then make multiple plans based on those parameters without showing the planning agent the code base, and then have a few worktrees to test the initial plans and not let the tests architecture decisions

Hello, does anyone here have any experience of building software factories? I know i missed the ATW episode but it talks about using Open Inspect (https://github.com/ColeMurray/background-agents) but there are loads of these open source agent orchestration repos and i can't figure out which one to trust. Also people are building factories with closed source like grokbot + cursor cloud agents / projects - https://x.com/kunchenguid/status/2090463366762676732?s=20. Does anyone have experience with either of these? OS or cursor based? My ideal setup is grokbot+cursor projects but hate the closed nature of this. Hope it is ok to ask non Baml question here..

Hi I’m still learning coding is there a good pdf guide or yt video on what baml is

also heyoo give me benchmark.. and use mojo utilies gpu more

also bad naming, similar to yaml or toml its baml and could face the fancy question of is html a programming language

this really need to backed up with some fancy companies...

heyoooo its not battle tested language right? there could memory leak or core level breaks etc... so, why should i switch?

hi all

I think Jev and LLM+BAML can be complementary. My "on-deck" experiment with Jev is to pair eval. Sort of council. LLM+BAML to give structured output akin to Jev's pritives and see if they agree or disagree. If they disagree, then flag for human review.

I view BAML as something like the glue that allows different various modern AI to fit with each other.

LLMs are amazing but latency is fundamentally very difficult. Stuff like Jev and the similar work being done in robotics are really good at realtime tasks that LLMs are too slow for. The real benefits come from clean integration which is something that BAML solves in super thoughtful ways. To me Jev and BAML simply aren't competitors, BAML is what allows you to get the most out of something like Jev and integrate different technologies together whether deterministic or not.

I was going to say, while I have not been utilizing BAML heavily, I did some poking around on the model side with needle3 and it feels like Jev ≅ needle3 + BAML taking the perspective of a web app developer/ux, I want to "just sideload a small LLM helper" into the SPA. (i.e. in cyberchef I could say "write me a recipe for splitting these lines, base64 decode, then comma separate by whitespace" and the sidecar "just does it" without everyone needing to host their own inference backends)

there is actually some true novelty in jev! the cost and speed of the model is quite quite nice. but that said, we do try and make it "just work" with baml. here's an example of something built using BAML + Jev: https://github.com/boundaryml/feelings/ But we're gonna dive a bit deeper during our podcast tmrw at 10:15 am PT: https://luma.com/all-about-jev

my newsfeed has been blowing up about Jev all weekend. I've not done a deep dive on it, so I'm sure I'm missing something, but I kept thinking "this whole product sounds like ~20 lines of baml" lol.

theo bringing up baml while talking abt jev!

my thoughts on UC: - Orchestration layer decision maker. -- e.g. should iuse this tool? -- would it be safe to run this tool? -- should I stop the loops? - Vibe evals or secondary judge

i am on typesafe api already how would i be able to use official baml with it so i can benchmark it for my current prod workflows

it does, I pointed claude to github.com/BoundaryML/feelings and was able to get it to write a client that went through Cloudflare AI gateway, since TypeSafe is on waitlist. Nightly BAML has TypeSafe client already built in

Does baml support jev model yet ?

Has anyone started playing around with BAML on Cloudflare lately? I noticed that the Cloudflare Workers data cap got lifted 2 weeks ago, which was one of the blockers on BAML on the edge. Claude tells me it doesn't even need to ship the WASM for it

Do you guys have a guide to of integrating human layer and linear

input = text / json text output = answers simple questions of three flavors - score based on rubric 0-100 - yes no with confidence score - pick item out of a list of options extremely low latency compared to llm structured outputs. answers many questions against same input in parallel

guys what is jev in layman’s terms + the usecase for it

😂

I also have replaced 100% of my Spotify and CVS usage with jev.

That statement seems a little hyperbolic

what does this smart guy mean? https://cdn.discordapp.com/attachments/1119375594984050779/1550944445489356941/image0.jpg?ex=6ab02d12&is=6aaedb92&hm=e7804023ff216e16ff51ed32eb10b41777a32908ebdc95e07be96b5903f8dc5d&