In today’s rapidly evolving technological landscape, Law Enforcement Agencies (LEAs) in India have a unique opportunity to augment their capabilities. Addressing this need, we propose the Adaptive Legal Analysis Engine. A novel solution integrates cutting-edge large language models and custom training methodologies to support Investigation Officers (IOs) in interpreting and applying legal frameworks. Central to the engine’s functionality is its custom training, with low-rank adaptation and quantization enabling efficient utilization of limited resources while ensuring accuracy and comprehensive coverage of relevant laws and facts. Also, vector embedding and vector databases are used to enhance performance of the model, capturing semantic relationships and contextual nuances within legal texts. The engine facilitates semantic understanding, similarity analysis, and contextualized recommendations by encoding legal provisions into high-dimensional vector spaces. The 4-bit quantization reduces the size of the model from 14 GB to 5.3 GB, and low-rank decomposition of gradient matrix reduces the RAM requirements of the model during fine-tuning. This paper comprehensively explains the development of the engine, detailing its architecture, training methodologies, integration of vector databases, and practical applications within the domain of Indian law enforcement.