I like your website but what you are describing is not a deep learning algorithm. I'm not a big fan of people who do something well and then start going overboard with buzz words.
You are most likely have a system that consumes content, compares that content against known hashed variants. If there is no match, you diff against known variants and check if the output matches any of the 'minimal' implementations.
If you can't match anything, you simply stage that content for a manual review.
Thanks for telling us what we do at Whatruns. I’ll let the team know ;)
On a serious note, I'm with you on how new start-ups go overboard with buzz words. As for WhatRuns, it was intentional that we do not use any jargons to advertise our product on the website, Product Hunt or HackerNews, so that it does not lose its charm.
For a new startup to achieve this scale in technology identification and accuracy compared to established players with more than a decade of development (and data), it is self-evident that manual labour would not yield such a result. In fact, technology breakthroughs and an excellent technical team were the reason why we decided to give this shot in the first place.
We plan to publish a comparative study on our experience with the effectiveness and superior prediction quality of deep learning vs normal pattern identification on our blog (which we will soon move to Medium). Stay tuned! :)
You are most likely have a system that consumes content, compares that content against known hashed variants. If there is no match, you diff against known variants and check if the output matches any of the 'minimal' implementations.
If you can't match anything, you simply stage that content for a manual review.