After a lot of exploration I decided to limit the scope as much as possible and target home cooks and food bloggers, with a focus on recipes.
I spent a few weeks of my spare time trying to get the parsing of free-text recipes to work (essentially MFP's import from link). The results were decent, but I realized that if it's only 90% accurate I still need a lot of user interface functions (like MFP) to let users edit/fix the inaccurate bits. It started smelling bad so I decided to strip it down to its most basic function.
I have no plans for now to use databases of branded products - the ones I saw were quite expensive and would be more useful for sites like MyFitnessPal, perhaps less useful for recipes/cooking.