The first project was for a large, (now) well-known fintech company. They needed to develop login integrations with consumer banks to acquire customer account information for verification purposes. But many such banks didn't particularly want to grant them any special API access. More importantly, these banks typically forbid scraping and made it explicitly difficult by implementing JavaScript-based computational measures required on the client in order to successfully login. I helped this company develop methodologies for bypassing the anti-scraping measures on several banking websites. However, I stopped working on this because 1) I felt uncomfortable with the cavalier way they were ignoring banks' refusals, then using the reversed integrations and onboarded customers as a bargaining chip for more formal partnerships, and 2) performing huge amounts of analytics on customer data acquired as part of the account verification process.
The second project was for a tech startup working on insurance and credit analytics. This company is one of several that popped up in recent years to use machine learning and social data in order to develop a more "complete" credit score (in their eyes). They had an impressive team of machine learning researchers but their data acquisition team was comparatively mediocre. So I worked with them to improve their acquisition methodologies for a variety of social media websites. I stopped working with them for three reasons: 1) fundamentally, I lost faith that their product was actually generating a meaningful signal over traditional means, 2) I was worried that the data they were collecting might introduce spurious correlations or illegal biases, and 3) if any team was going to do this correctly, I didn't think this particular team was the qualified one to do it.