atcole
atcole on Hacker News
It can be done, working on it right now :) I would say the hard part is validating that it's correct (you need a corpus of verified returns), and the actual act of filing still requires manual effort.
My initial runs though show that you need a medium beefy setup to be able to get accurate results. Anything less than Gemma4 31b and you get some input document hallucinations.
https://github.com/captaincole/OpenReturn
Yeah! So there are two levels to my validation.
1) I've built the engine and tested it based on CPA reviewed real returns. You can see that in the scenario tests
2) I also have a mechanism for people to test the engine against their previous returns privately and then send bug reports of something wrong
The engine currently only works for 2025 because the 2026 documents aren't out yet from the irs.
There is also definitely a place as well in the tax world for general estimates. For example calculating the tax consequences of a stock sale or iso exercise can get really complicated without a system and cost $800+ with a CPA to do.
I think it will be possible to do with the right harness in the future. I've tested gemma4 31b and it's pretty close with an engine backing it.
I am. For free too! open source and local model capable (albeit most small models and hardware setups can't do it yet). I have some evals provided on document ingestion.
https://github.com/captaincole/OpenReturn
Check it out
This is great that this is out there. I've been working on making an agent based experience that helps you put together your whole return or make an estimated return based on future assumptions and solving for the federal tax code has been a very token expensive operation!
https://github.com/captaincole/OpenReturn - Still in Beta
I'm going through their engine as well and validating mine. My engine is built to include the CA tax code also.
The biggest reason that turbo tax and the other companies survive though is that in order to allow efile you have to be a registered agent and that process is crazy expensive and difficult.
I'm working on an open source agent and engine that can calculate your Federal and California taxes. My goal is to build an AI system that can act as your CPA, and hopefully allow you to do super cheap tax estimates and planning.
Looking for help as well!
https://github.com/captaincole/OpenReturn
Hey, I'm working on something similar here:
https://github.com/captaincole/OpenReturn
And I have a concept as well for how people can verify the solver by running their own previous returns through the engine. Would love to collaborate as this is a side project for me as well.
Just this morning I added several copies of books 1,2 & 3 of the Wheel Of Time series to a random Little Free Library because I had so much fun reading them and I hoped others would as well.
Hoping this continues.
Posted by
atcole on Friday January 25, 2019 @01:19PM
from the blog.forgeglobal.com dept.
Posted by
atcole on Sunday February 18, 2018 @02:29PM
from the medium.com dept.
As a qualifier, missing by 30 seconds on a 85 second lap is not very close.
But a 1:55 at Thunderhill would be consider a fairly decent amateur time on a 600cc bike so they are definitely making progress.
In cases similar to desktops with electron vs. native (although with different requirements) where either you need access to significant native features that are not supported in the browser, or if your app is process intensive, i.e. graphics heavy games.
The statistics shown on how helmet laws actually lead to an increase in cycling accidents is very interesting. I wonder what the correlation is on this? Is it because cyclists not wearing helmets generally bike in a safer manner? Or maybe if there are no helmet laws, then streets designed for bikes are designed with more safety in mind?
An interesting note about bike shares usage. I have seen some very effective bike sharing programs but also some very ineffective ones. In San Jose, there is a bike share program, but the bikes themselves cost $9 to ride, while in London the bike share cost 2p (~3-4$). The price point made all the difference to me.
I think the telling quote of the article is the last sentance which sums up his feelings on the current US situation
- "The situation that many in-demand U.S. cities find themselves in is an example of extreme “market failure,” which is what bureaucrats call regulatory failure; and is also extremely unnatural"
Basically he is saying that "bureaucrats" ineffectively regulated the market to the point of failure, or where the market could no longer compensate for the demand which causes a lack of affordable housing.
So I think he isn't calling for more / less regulation, but better slash more effective regulation that allows the market to operate.
His proposal is two-fold, for those who want a summary.
1) Build better transit to connect areas that are affordable (or areas at all i.e. San Jose / Oakland) to central business districts
2) To accomplish this by creating a centralized governing authority for the whole Bay Area.
I wonder if this kind of consolidation has ever happened before. Did New York City absorb areas that it now controls as it evolved, or did the parties agree afterward to some symbiotic relationship?
I know we will probably never get to see this (or at least not in the near future) but I would love to see the cost to reuse.
Thank you. I'm glad someone corrected this man's outrageous claim that we would let anyone in for less than 9 bags.
This is a fairly technical analysis, and the terminology used in many cases is above what I know about networking. But the one quote that will stick is this.
"There is no such a thing as a temporary change or workaround: In most cases, workarounds are tech debt."
This headline is a bit misleading. It says that Tests of the EMDrive were successful in the laboratory setting and that they are currently testing it onboard the Tiangong 2.
It does not reference results of the Tiangong 2 tests.
I navigated to the IP address for Github, but it still tries to resolve to github.com
Instead of using metime, I actually use rescuetime which is a fairly similar application if you also track your offline time.