I've worked on a Jai Alai handicapping program for a few years now, on and off. I'm in Orlando, so I'll only work on it during the season, and then put it away for the rest of the year. I run it in MATLAB btw. The main part of the game is very simple to program, but I spent a long long time coding in all of the tiebreak situations needed to give me good trifecta odds.
I haven't got nearly as far as you yet as far as pulling in past data and using statistics, parameter estimation, etc. I use Monte Carlo simulation based on parameters I enter for a given game, basically a strength rating for each player. Compare the strength ratings of two opposing players in a point during the simulation, and you get a probability of each player winning that point.
I usually bring my laptop with me, and do it all on the fly, so if a player is hot, or a player is cold, I can take that into consideration - although for the most part my ratings for each player that I've watched for years stays basically the same. It takes a couple minutes for the simulation to run to a decent level of convergence (a million sims), and then I get a couple minutes to enter my picks. Works alright, like you said before, probably better than the average bettor, but still lots of room for improvement.
