Really look at it... pore over it... think it... be it... once you start dreaming about price and volume data, you'll know you're on the right track...
...until you think you might notice an edge or something somewhere... then write unit tests to see if that hunch about the pattern your monkey brain thinks it saw is actually consistently profitable.
Then you do all of the real work that is implementing the actual trade execution engine. (Yes, I just yada yada'd over months of work)
That's how I went about it at least and it's working out great so far!