Less big data, more ML: you could build a rudimentary shazam, for either audio or images. Use small samples (~1sec clips, or 10~20% image crops) to match against an existing corpus. Bonus if the audio is distorted &/ the image is skewed/rotated a slight amount.
Not as fun: write a k-lambda [1] interpreter in x64 or ARM. More fun: write it in Python or Go. Not fun at all: write it in Verilog/VHDL.
Big data: I hope you like football! Just kidding, there are other datasets, too. You could simply create a responsive website for exploring the dataset. Or do some interesting analysis. Up to you. [2], [3]
Not as fun: write a k-lambda [1] interpreter in x64 or ARM. More fun: write it in Python or Go. Not fun at all: write it in Verilog/VHDL.
Big data: I hope you like football! Just kidding, there are other datasets, too. You could simply create a responsive website for exploring the dataset. Or do some interesting analysis. Up to you. [2], [3]
[1] http://www.shenlanguage.org/learn-shen/shendoc.htm#The%20Pri...
[2] [large!] http://aragorn.org/data/sports/NFL/seasons/metadata_since_19...
[3] http://www.wired.com/2012/09/nfl-momentum/