Good To Know—AI Assistant for Understanding Legal Documents

• 

  Good to Know

Inspiration

We all had situations in life facing down a legal document and wanting to know in plain language what each clause and paragraph meant to me personally; facing the decision of signing the document, asking for changes, or walking away from the deal.

What it does

The primary MVP for today puts the text of a short (2-20 pgs) legal document into a Bedrock prompt and returns a ranked list of potential issues spotted in the document, sorted by priority of legal risk. It has an optional feature to do A/B/C testing across a secondary and tertiary LLM, returning the results so the user can choose which LLM best serves the needs of that document. We have also written preliminary code to provide a friendly user interface able to test between more than one usable LLM.

How we built it

We focused our efforts on the prompt engineering for Bedrock to return the ranked list of potential issues the LLM has spotted. The objective was to get a prompt that returned good-enough results, i.e., things that might be good to know without attempting to be legal analysis.

Challenges we ran into

Getting the full dev environment running in enough time to code our interface, due to the effects of the team coming together over the first few hours.

Accomplishments that we're proud of

Creating an idea and MVP for a product we know we'd use and the market really needs.

What we learned

Find a way to get the technical configuration and organization done early in the hackathon so there is more time to code.

What's next for Good To Know

Pitch, pitch, pitch!

Built With

• bedrock
• convex
• launchdarkly
• python