It was interesting. I watched it a couple of times and I have to say his approach was pretty good. His problem was his knowledge of horse racing though. He took the domain, assumed he had enough knowledge, and tried to analyse as much data as he could which wasn't enough. He then tried to come up with conclusions based on insufficient data by using "Machine learning techniques" but that really doesn't work.

Actually I did like the bit about "not all candidates are the same in the gene pool. Some are stronger, more immune to disease". This made sense when you consider the odds are the strength of a horse and some negative factors are not as bad against some favorites are they are against long shots.

The best line came at the end of the video in answer to a question. the question was "Do you currently bet this?"
His answer was "No its illegal to bet on hong kong racing in the USA."

That was just an excuse. if it was that good at prediction he could have flown to Hong Kong and bet there and easily recovered his money. He could have sold the software to a syndicate, he could have moved to a more bet friendly country. He had no confidence in it because he didn't understand it. It was a black box to him on why it might work and also why it does not work. He can't fix it.

If he spends more time on it (which hopefully he has) then he could fine tune it and use it as an input to a betting system.