Custom LLM Fine-Tuning — Indian Legal Q&A

A custom LLM fine-tuned using LoRA adapters, purpose-built for Indian legal reasoning — trained on statutes, the Constitution, and real court case data to deliver accurate, explainable answers to legal questions. Deployed on cloud infrastructure running the fine-tuned weights — no third-party LLM API dependency, full data sovereignty, no vendor lock-in. End-to-end build: data pipeline, fine-tuning, iterative retraining, evaluation harness, and a working demo.

Key details:

- LoRA adapter fine-tuning of an open-source reasoning LLM for a specialized legal domain

- Deployed on cloud infra running custom fine-tuned weights — no third-party LLM API dependency, no vendor lock-in

- Data pipeline merges synthetic chain-of-thought examples with real judicial case data and constitutional/statutory text

- Chain-of-thought reasoning for explainable, auditable legal answers

- Iteratively retrained across multiple checkpoints, validated with a baseline-vs-fine-tuned evaluation harness

- Interactive chat demo for walkthroughs