The situation
The law firm's lawyers spent hours reviewing contracts and NDAs by hand. Client confidentiality ruled out sending documents to external AI services, so cloud LLMs weren't an option.
What we did
We deployed Llama 3.1 70B locally on NVIDIA A100 hardware, served with vLLM and a FastAPI interface, to analyze contracts and extract key terms and obligations. All processing stays inside the firm's own infrastructure.
The outcome
Contract review now takes 5 minutes instead of 3 hours, key terms are extracted with 98% accuracy, and client data never leaves the firm.