I'm a system architect and I work as a presale-consultant for the customers. Basically, my company developed a custom solution based on Hadoop+Hive+Hunk, plus several units doing machine learning, mostly kohonen networks, bayesian and FFP networks, to deploy as jobs. What now is named "big data", which was named as "supercomputing" like 25 years ago. The customer tells us what he needs to know, starting with their data, and we try to implement.