SCI 2026ACCEPTED
LLM-Based Courtroom Simulation for Indian Civil Law
AUTHOR // KUSHAGRA GOYAL·NISQ & SYSTEM LABS
INFORMATION
- TypeResearch Paper
- VenueSCI 2026
- Index StatusACCEPTED
CONTRIBUTIONS
- Multi-agent state machine modeling the adversarial legal process.
- Contextual retrieval template specifically tuned for Indian Civil Procedure Code (CPC).
- Rulings evaluated by legal experts with high concordance scores to actual judicial precedents.
ABSTRACT.
An adversarial multi-agent RAG (Retrieval-Augmented Generation) framework designed to simulate Indian civil law courtroom proceedings. The system employs specialized agents modeling judges, defense lawyers, and prosecution lawyers, utilizing a vectorized index of the Indian Kanoon corpus. The agents dynamically cross-examine evidence, argue legal precedents, and draft judicial rulings aligned with Indian constitutional law.
METHODOLOGY.
Constructed using LangGraph for multi-agent choreography. The legal documents are chunked and embedded via HuggingFace models, indexed in Pinecone, and dynamically queried during agent turns to generate contextually accurate arguments.