It’s been about two years since I left full time criminal defence practice to ‘build a legal tech company’. I’ve learned a lot along the way, and have built some cool and useful things, but I do not have a revenue-producing company, and those around me are getting impatient.
So after a long conversation with my wife, we’ve decided to put a time limit on it.
I have 90 days to hunker down and build something that has real commercial potential, and perhaps even could attract funding. (not that there’s much of that flowing around nowadays).
Failing that, it’s time to get a real job.
I’m going to apply what I’ve learned to try and maximize my chances:
-
it should not use government data. I should have listened about this originally, frankly. Tussling with the government is not a small fight, even when done for righteous reasons, and requires more experience and funding than I currently have
-
it should be something people immediately want to pay me for. No more ‘but it should exist, or “it’s useful once you learn how to use it”. No no. Wow or go home
-
it should be simple to explain, and simple to use
-
it should have real growth potential
-
put my head down and do all the right things. That means shamelessly stick it in front of, and chase, everyone for use and feedback, while constantly iterating
-
Put everything else aside.
Given all of that, I’ve decided to hop on board the generative AI train and stick a chatbot on top of areas of criminal law that I know from practice people need everyday, and then see what the reception to that is.
If I can build something that even Ontario criminal lawyers want to pay for, then I’ve got a shot.
Where to start? The following:
-
adapt my code calls to the ChatGPT api
-
experiment with LlamaIndex and Langchain to see which is the easiest way to get started effectively (or perhaps the combination)
-
find the easiest vector store/search method and incorporate
-
pick one or two high-needs areas that criminal lawyers would want to have easy and quick answers to now, and start there. The smaller and more important the area the better
Ideas for areas to start?
-
impaireds
-
sex assault applications (276/278)
-
drugs
-
evidence
Types of questions you should be able to ask?
-
offence elements
-
application procedure
-
common disclosure items
We’re going to break this down into do-able chunks, and then execute.
Chunk 1 - query a recent impaired textbook using the simplest of the above options, see what we’re working with