Strengths: Developing a mechanistic model extending the SIR framework to fit multiple waves of COVID is not so easy, and this project makes good progress on that. The over-dispersed measurement model combined with the time-varying transmission rate is sufficient to obtain a model that is competitive with an ARMA benchmark comparison.
Points for consideration:
It is a reasonable assessment that the strong weekly pattern may be the missing piece for getting a mechanistic model to beat an ARMA benchmark.
Over-dispersed process noise might also help with this.
A time-varying measurement model would also make sense in the context of COVID.
These suggestions go beyond expectations for a final project: this is a nicely worked project that makes substantial progress toward its goal.