Court Judgement Drafting: Building QnAs, copilot response cycle and final merger

Building QnAs
- Providing Document’s sections and semantics as “Context” in our few shot prompt
- Giving examples of few questions for a section in the few shot prompts
- In this way the LLM will generate enough questions pairs on a every document section provided.
- We can also set no. of questions, the tone of questions, length and other requirements.
Building the interface + copilot response cycle
- Rounds of question and answer session with client will occur.
- If sufficient info is not recieved then copilot will again ask a better question to get all the details needed.
- After recieving all the info then the document drafting process will begin.
Drafting the document
- Here we will merge the pertinent document sections.
- The merging will occur in a defined layout or structure according to the need.
- This will result in our final draft of the required document.

Goals

Input: Schema provided by step 1.
Create questions whose answers will add details to the values of the schema.
If values cannot be extracted in the required detail from the user answer, ask back more questions to the user.
Incorporate different kinds of datastores like frozen values, unique elements and common elements.

No response

No response

No response

Ask LLM to create questions:
- LLM evaluates if the user answer contains the required answer.
- If yes, ask the next question.
- If no, then create another question that answers the question.
Create a decision (only works if you know the options based on which you predict).

No response

Court Judgement Drafter

SamagraX

⁠Service Delivery

Machine Learning, Natural Language Processing, Python

@ChakshuGautam @GautamR-Samagra