Someone mentioned Enthought. Kitware and Galois are two other prominent examples. Kiware operates on collaborative development and grant subcontracts, and propr

Someone mentioned Enthought. Kitware and Galois are two other prominent examples. Kiware operates on collaborative development and grant subcontracts, and proprietary product dev. Enthought seems to do high-end consulting enabled by rapid development in Python. I've only worked with their (impressively) high-quality libs. It's not hard to imagine they have ~one order or magnitude competitive advantage over companies doing similar work in C++ (reminding me of PGs essay about choosing lisp for viaweb) The performance crunch for their Python apps is really in the underlying C or C++ libraries -- which are open-sourced so they don't have to maintain it all themselves. One thing to note about KW and ET is that they build libraries and invest in leveraging tools to improve productivity (KW: CMake, CDash, etc. ET: Traits, Chaco, etc.) One approach to try to emulate KW/ET: get an SBIR or build a small team around grant subcontracts. For this you need connections, recognizable core technical prowess (VTK, NumPy), and fantastic recruiting instincts. And it scales very slowly (KW: 13+ years to 100 employees). Another way is niche/boutique consulting, but this generally relies on high-end skills -- which it sounds like you have, but I think the problem here is that $YOU don't scale, so growth potential is more limited. It's also harder to scale boutique dev. because it will very likely be covered by NDAs in many industries, so you are essentially stuck as a contractor. I'm not sure how viable consulting for a large med device company (GE, Phillips, Siemens) would be long-term, because anything you can do, they can throw money at and repli Personally, I'd love to see something that fills the gap between Excel and MATLAB (depending on your field, you may be shocked to learn that people do Real Science in Excel every day; I still find it shocking).. The market is potentially "big enough", but the business problem is that technical competence seems to be binary and the 0s have excel already and are afraid of change while the 1s have MATLAB or SciPy. To bridge the gap you need software that "just works" in a simple way, a la Dropbox. This isn't necessarily a "hard" or interesting scientific problem, it's much more about design and engineering. For examples of some steps in the right direction, have a look at KNIME, RapidMiner, Orange, and VisTrails,