My last days at the ArticuLab got rather hectic rather quickly--I think mostly because I hadn't accounted for the InMind debugging and demos and switched gears between that and my own project/offboarding quite a few times. But it was a good kind of busy--I only wish I'd been there for more collaboration on the overall project earlier in the summer (also that the demo on Friday went more smoothly; more on that in the Research section).
The End (or rather, Chapter Fourteen)
After fourteen incredible weeks with incredible people and adventures, the idea of leaving the ArticuLab doesn't even seem possible. I wrote a blurb about all of this (i.e. a recap of my summer, now that I've lived it) here.
Florian very kindly offered and dedicated time to help me rehearse for my final presentation; I completely botched the rehearsals, but the actual presentation wasn't a total flop (aka the success is thanks to Florian).
I wasn't particularly useful at InMind debugging, but I tried my best: mainly a) parsing log messages and organizing them into CSVs and b) upgrading our NLG to account for users' past utterances/preferences (so if someone said "I like Carrie Fisher," our movie agent could respond with "Carrie Fisher is my favorite too").
One thing that I definitely should've done better: integrating my incremental changes into the InMind system. Because I branched off of the original multiuser framework (MUF) repository light-years ago and we're using a different NLU/DM repository than before, I actually ended up newly branching off the current MUF and NLU/DM respositories and manually transferring my changes (badly done, Viv). Fingers crossed that'll all work out in the end.
Our InMind team met in the conference room just about every day to tweak and run code (aka copy/paste and make faces at a lot of error messages).
On both Wednesday and Friday, Justine came in to test out the movie agent. In other words: we explained any new features; she chatted with Sara (which is what we're still calling the agent at this point); we talked about how what was supposed to happen during these conversations didn't quite happen; and everyone established a to-do list for the next month or so. Oh, and Justine brought food, too.
Gave my final presentation! After hearing most of my fellow interns (all of my DREU friends, plus a handful more) deliver theirs over the past month or so, mine felt more than a bit surreal. Lab meeting was small this week (some of the usual lab members were missing, as were the interns that have already left), but in a good (half cozy, half the-less-people-to-embarrass-myself-in-front-of-the-better) way. And everything ran smoothly: I remembered to say what I needed to say and finished before the ten-minute time limit, and people had lots of curious questions at the end.
These are more or less lifted from Bitbucket, where I keep a README of these things.
Ran user study; mtg with Florian; presentation; debugging session.
Checked stats; presentation; phone code; altered analysis tool; debugging session.
Phone code; debugging session; NLG changes; demo.
NLG changes; presentation.
Lab mtg/presentation; incrementality; organizing data/offboarding; demo.
Hot Sausage and Mustard*
Jingya, Wei, Florian, and I got food truck food for Tuesday's lunch (something we've talked about doing for eons), which was a fun change from our usual routine.
* it was actually Indian food, but I couldn't resist quoting