Really cool - even cooler if some farming related hardware on a designated plot of land can be setup so it's more than an ai agent finding someone to hire via apis.
The upload often doesn't work with redacted information because redacting sometimes changes the pdf's structure and can make the text extraction less reliable. I'm working on making the pdf processing more resilient.
I am not dismissing the concern, I was stating the tool solves an even larger concern. I'm doing everything I can to setup it up to be secure, private, and worthy of trust and addressing the feedback points.
If you have suggestions more than "don't trust this random internet tool even if it gives you free advice, regardless of the value it offers", please let me know [thanks emoji]
AI tools like this don't generally need their own models or fine tuned LLMs. Uploading mortgage documents to ChatGPT may get some useful results but it really won't be comprehensive, it most probably won't take into account jurisdictional laws and mortgages rate data, and would be hard(er) to trust because of hallucinations without guardrails.
Technically this is an LLM wrapper but it utilizes around 5-10 prompts to extract and structure data from the document and ~15 prompts to analyze.
Extract all the text from the PDF, turn the pdf into images, send the text for each page along with the image to an LLM with a desired output strucutre.
hacker news, reddit, and other communities. Hopefully there will be a strong incentive in the future for real estate agents or mortgage brokers to recommend the tool. Also I think the name closing.wtf will be good for distribution.
The tool analyzes mortgage loan estimates and closing disclosures to find every potential place a buyer might be overpaying, every negotiable fee, and show the buyer a comprehensive AI analysis so they don't get screwed over and can have peace of mind on buying their home.
When I was closing on a house, I called a few friends to help review my mortgage, and we found lots of mistakes. For example, I was getting charged transfer tax, which didn’t make sense for Florida where the seller typically pays that. The deeper I dug, the more I realized how much gray area there is with these documents - what’s negotiable, what’s inflated, what's normal in the property's jurisdiction, and what’s just non-competitive but seems ok since there's a lot of simplified complexity that goes into mortgages and what's an extra 1-3% on a mortgage that's "just going to be refinanced in 6 months" when interest rates go down.
The analysis is generated via ~15 prompts and the output is pretty useful if you're going through buying a property with a mortgage.
The goal is to help mortgage borrowers get better deals on their mortgage. All mortgage terms are outlined in the loan estimate and closing disclosure documents as required by the CFPB (consumer financial protection bureau) - https://www.consumerfinance.gov/owning-a-home/loan-estimate
The tool requires uploading these documents since they have all the data needed to generate the analysis on the mortgage. Technically I can have the user paste in all the data via a textbox, but then it would be much more difficult to reliably parse. Also requiring the upload of the original documents is a much cleaner experience.