[complainic]

I'm a lawyer turned programmer. Generally agree with the above but would add:

   - NLP applications, many practitioners and judges may not trust the software especially if implemented by a hobbyist and prefer many paralegals review manually.  I'd pursue this more out of interest or to solve a personal workflow issue but be wary of generalising unless it becomes your main focus.
   - Integrations with word/excel can definitely be useful. I wrote a bunch of simple VBA add-ons in Excel to help with mundane parts of managing a large e-discovery. But I found it a real slog to learn the word/excel specific stuff, so wouldn't recommend it as a first hobbyist project.
   - Learning how to ingest various docs and do regex saved me the most time. If you do a lot of e-discovery and your firm's platform of choice supports regex just learning that grammar is probably the best bang for buck.  Downloading and searching a smaller number if files locally can also be good. Was always disappointed when the files I wanted were in a "secure" dealroom so I couldn't legitimately download or scrape them.

[lawtalkinghuman]

On NLP, in the UK there has been one miscarriage of justice (a wrongfully convicted man spent four years in prison) due to mishandling of text messages by police and prosecutors, and some close misses. Because of the time pressures on police and prosecutors, they were just looking for a bunch of keywords unsystematically rather than reading the material fully, and failing to disclose relevant material to defence lawyers.

See: https://www.independent.co.uk/news/uk/crime/rape-conviction-...

Getting text automation software wrong can have pretty horrible consequences. AI/ML tools should be an aid to proper review of documents, not a replacement for it.

[osdotsystem]

Thanks for your input!