I wanted to know in what ways (with examples if possible) can dummy variables violate OLS assumptions?
Dummy variables on their own, being Boolean, are not violating the OLS assumptions.
However, if you do not drop one of them, then you would fall into the dummy variable trap. It is explained in the lectures, but in short, if you have N dummies which mimic a given variable, you can explain it with N-1 of them. Therefore the Nth dummy is useless. Moreover, it is perfectly explained by the other N-1 dummies which introduces perfect multicollinearity to the model. And this violates the OLS assumptions!
The 365 Team