Main feedback: great work!! i especially liked slide 10 on the pre versus post preprocessing contrast, the illustrations of the sentiment score as well as updating the sentiment lexicon, and the over-time plots. This is definitely in good shape already so no pressure to do a ton more, but a few comments on small things

• Money features: as discussed, would try to check robustness (if time) to extracting monetary amounts based on either regex like dollar symbols or if there's a currency extraction since i wonder if high correlation on slide 16 is artifact fo the filtering to range specified on slide 15
• slide 18: based on this graph, it seems like there's a high frequency of tweets that (1) get 0 engagement and (2) that get weighted to zero so effectively dropped from the prediction model - i think this makes sense but just being clear then the analytic sample is tweets with a non-zero amount of viewer engagement
• slide 19: if there's an equation representation of the bitcoin differential value that'd be helpful (as someone unfamiliar w/ this topic is it difference between day t open and close price, close price on day t versus close price on day t+1 etc)
• slide 27: maybe illustrating which months are in training, validation, and test (or days/weeks depending on the unit of time used to split)
• I know we rushed through the last set of slides but related to the above about the dependent variable, it seems like you could: (1) use the return as the DV/Y even in the ML model, since it seems good to try to predict day to day changes rather than price on a fixed day given high correlation; (2) i think you'd want to use the sentiment on day t (so not the change in sentiment but whether sentiment on March 8 2021 predicts change in price between closing price March 8th 2021 and closing price March 9th 2021)
• discussion of timestamp of tweet related to time zones/opening and closing of markets - writing out the above comment made me realize that say the markets close at 4 pm EST--- you'd want to the day $t$ tweets to be posted before 4 pm EST and then everything after market close goes into the next day--- so when merging the tweets to the price data, making sure the tweets are attached to pre market close hours