Hey friends, some update:
This is my demo:

It uses RNN to classify 3 categories of laws: nuisance, dangerous driving, work injuries. RNN is trained on law cases (30 + 30 + 20 = total 80 cases, respectively). The performance is quite good (even surprised me a bit), but still it's not very useful for real scenarios at this stage.
So I propose a model for "fine classification":
but it's not implemented yet. The 2nd stage is more challenging.
There're some quarrels in the team and I'm waiting for the organizers to mediate. (I just talked to them and there's some chance of reconciling with the team)
In the meantime I joined another team which is similar and they entered the HK Law Society hackathon. So basically we're still open to recruit more partners 😄
See if you're interested ☺
PS: I used the word vectors from FastText and it's great. But we didn't call FastText, just used the data file (2 Gb size, 1 million words).
YKY