Hey, nipnipj. Quite an impressive knot of hair (I cannot not see it) you plotted there. It seems like you are knowledgeable with Pymc3’s methodology, so the next step is to think about what kind of model would best fit Y_obs.
Have you thought about what distributions may better fit your data? The T-distribution, some Gamma distribution, a mixture? Think about and play with these concepts for a bit and see how it goes.