I checked out that SVM link you provided, it seemed very interesting. I've messed around with neural networks in the past, with no real success with regards to algorithmic trading.
It seems to me that the hard problem is figuring out how to represent price data as "points in n-dimensional space", in a way that makes sense for the algorithm. I had the same problem with neural networks, figuring out how to translate prices into something a NN could use as meaningful input.
I'm a programmer by trade, but I have very little experience with AI so maybe I'm just dense... :doh:
As far as making our own trading software goes, it's a really big job. It the classic case where the potential users think it's easy because they (we) are very familiar with the problem domain, and can sum it up quickly when asked. But in reality, when you consider the myriad details it would be an almost impossible undertaking to make something with a small team in a reasonable period of time that does decent charting, indicators, analysis, data management and automated order entry.
There's a reason why all the commercial packages seem to fall short, and it's not because they all don't know what they are doing. Rather, I'd blame short release cycles and pressure from competition to keep up with the latest bullet point features. Not to mention differing ideas of what is most important in a software package.
Although I do sort of suspect that the financial industry tends to have sub-par programming for some reason. Perhaps because the programmers are hired and managed by finance MBAs and not technical folks. No offense to any programmers or MBAs out there, it's a just general impression.
For me the most important thing is that I want to be able to code my strategies in C# or some other "real" language, rather than some kind of brain-dead "easy" scripting language.