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

[AryanPrakhar]

Ticket Contents

1. 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.

2. 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.

3. 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 & Mid-Point Milestone

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.

Setup/Installation

No response

Expected Outcome

No response

Acceptance Criteria

No response

Implementation Details

1. 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.

2. Create a decision (only works if you know the options based on which you predict).

Mockups/Wireframes

No response

Product Name

Court Judgement Drafter

Organisation Name

SamagraX

Domain

⁠Service Delivery

Tech Skills Needed

Machine Learning, Natural Language Processing, Python

Mentor(s)

@ChakshuGautam @GautamR-Samagra

Category

Machine Learning

[AryanPrakhar]

Weekly Learnings & Updates

Week 1

Week 2

Week 3

Week 4