Legal work is famously slow and expensive, much of it spent processing documents. LegalMind is an “agentic” AI, one that can carry out multi-step tasks, built to automate parts of legal workflows and bring costs down.
The problem
The legal industry struggles with inefficiency, high costs and manual workflows. Traditional AI tools lack the adaptability to optimise legal operations end to end.
What the research did
We built LegalMind on DeepSeek R1, an open large language model with strong reasoning abilities, and fine-tuned it on a structured legal dataset. As an agentic system, it can plan and carry out sequences of legal process tasks rather than answer one question at a time.
What we found
Compared with baseline AI models, LegalMind reduced costs by 42.6% and improved document processing speed by 60.8%. Scalability tests showed it could handle 100,000 queries efficiently. Case studies in law firms, corporate legal departments and government agencies showed reductions in case preparation time and operational costs.
Who benefits
- Law firms and in-house legal teams
- Government agencies handling large volumes of legal documents
- The public, if lower costs make legal help more affordable
Limits and what comes next
Legal AI must be supervised by qualified lawyers, as errors can have serious consequences. Confidentiality, accountability and different legal systems across countries all need careful handling.
The team
This research was led by N. V. D. S. S. V. Prasad Raju (BITS Pilani) with me, Nikhil Patel, Olivia-Roxana Alecsoiu, Priyabrata Thatoi (Amazon), Salem A. Alyami and AKM Azad.
Frequently asked questions
What is agentic AI?
AI that can plan and carry out a sequence of steps towards a goal, not just answer single questions.
How much can LegalMind save?
In experiments it cut costs by 42.6% and sped up document processing by 60.8% compared with baseline AI.
Can AI replace lawyers?
No. It can speed up routine work, but lawyers must review and take responsibility for legal decisions.