mattfrommars

Thanks for the reply. I lost you in this part, Great questions on scale: the whole way we designed our engine is that in the happy path, we actually use very little LLMs. The agent runs deterministically, only checking at various critical spots if anomalies occurred (if it does, we fallback to computer use to take it home)

I assume you send screenshot to claude for nest action to take, how are you able to reduce this exact step by working deterministically? What is the is deterministic part and how you figure it out?

Looks great to automate workload for Windows desktop application. I'd love to understand more deeply how your application works, so the set of commands your backend send is click, scroll, screenshot. Does it send command to say type character into an input field? How is it able to pin point a text field from a screenshot? Is LLM reliable to pin point x and y to click on a field?

Also, to have this run in a large scale, Does it become prohibitively expensive to run on daily basis on thousand of custom workflows? I assume this runs on the cloud.

That could be a good question for AI to answer.

Are we really shocked by this? It's screaming obvious but we still have many people whose head is buried deep in the sand.

Israel have claimed the title to start worlds first live stream genocide. It took them long time to get to where they are and know they will be forgiven.

It's crazy to think but is true, they have normalized killing of Palestinians.

Can this work on intel npu unit?

Anyone know how long does the context last for running model locally vs running via OpenAPI or Cursor? My understanding is the model that run on the cloud have much greater context window that what we can have running locally.

Tangent thought as I've been trying to adopt new programming language. I'm amazed how prolific is typescript in open source project space.

It's year 2025 and we have yet to have impact of CUDA like what Java had in the idea, "write once, run it anywhere"

Academia and companies continue to write proprietary code. Its as if we continue to write code for Adobe Flash or Microsoft Silverlight in year 2025.

Honestly, I don't mind as Nvidia shareholder.

Amazing stuff. I think I'm in the boat as you OP except I just can't get myself to commit to a project since 1) no idea if it works or is unique 2) look for better opportunity and spend time doing LC/apply for job.

There is only so many hours after work.

> Once you’ve validated your idea,

Could you really break this down. I feel you overlooked this part. How do you filter through tons of ideas into idea that for sure or has good chance for success. For example, to build a yet another crypto exchange is both super technical and far too regulated. What is a known strategy how founder/dev nail down and commit to a project?

If I understand OP journey, was he fortunate to have been scholarship to fund his studies to become a pilot?

I was looking into pilot school here and they cost upwards to $100k

Btw, there is also LocalSend. Pretty nifty app

Is this a native IOS app?

Isn't this what lovable and that tool provided by vercel?

Does anyone know how start up or project like these are built from ground up? I have very very command of react and I can write rest api using spring or express.js. I know finger too about prompt engineering. What does it take to make something like this?

OP, what is your background to be in the position to build this? I've been itching to build or have an idea to make something complicated but only true thing I kind of know is building API and integrating with React application.

How were you able to deduce complexity into workable item/task to take 'I will alternate to Heroku' into fully functional open source project?

I haven't really use Heroku, so to figure out 'feature to be implemented', I would start with pricing page. I am sure my approach is flawed here.

You should ask youself, would you pay for your own services?

$250 a month is still a lot of money in India to spend on a digital product.

I am envious of folks who get to use AI on daily basis to do productive tasks. My use is very limited to asking it to summarize things to asking to explain LC problems.

We use an internal 'no code' tool to do complicated data transformation logic. Our flow has become so large and convoluted, it is extremely difficult to put feature and debugging is absolute nightmare.

You have no idea how many times we had wished all this data transformation was in vanilla Java/Python. Our life would have been so much easier. But at this time, it is too large to ignore.

How do people go from web dev/gradle/java knowledge to build something like this.

For me, sometimes I wonder since I never did CS undergrad, I will never understand any of this no matter how many programming langauge I learn. I did dabble a bit with OPENLY, LIBGDX, GODOT, unity but to come to with cloth simulation from scratch, damn

What does this look so similar to v0 from Vercel?

Who comes with the UI UX design? Is it to show the $5.3 million dollar in action?

anyone had success using Gemma 3 QAT models on Ollama with cline? They just don't work as good compared Gemini 2.0 flash provided by API

WTF

You built in in one afternoon? I need to figure out these mythical abilities.

I've thought about this idea few weeks back but could not figure out how to implement it.

Amazing job OP

Which backend API provider are you using? Is there option to run this locally with self host LLM?

One person? What do you mean? It literally says in the wiki more than one.

This isn't solo dev game project.

I just find this very sickening. On one hand, Google laid him off and best excuse I can think of is, "reduce operational expense". On the other hand, they happily paid $32 billion dollar for Wiz where the both fonder net around $3 billion dollar.

I don't know guys but there needs to be some crackdown on this bull that is going on with corporate America. Something feels really off lately with these tech companies on how they make excuse and get away.

Isn't the idea of AI agent talking to each by telling LLM model to reply say in, JSON and with some parameter value map to, say function in Python code? That in retrospect, given context {prompt} to LLM will be able to call said function code?

Is this what 'calling' is?

I'm just curious for folks who have read through the material OP suggested as well as, the book linked in this HN thread, are your guys primary motivation to understand and fill in that curiosity part of your head vs making a career out of this?

Is it reasonable to think that if one grinds to the book suggested here and background in web/dev SWE, one can break into ML/AI role?

is this a native mac app? want to know for curiosity sake. i am new to mac development, i inspected in application and it looks like python qt? if so, why this and not say electron based?