⚖️ LegalTechOn Time📋 Fixed Price

AI-Powered Document Intelligence — UK Law Firm

UK6 min readUpdated June 2025
Region
UK
Contract
Fixed Price
Tech Stack
9 Technologies
IP
100% transferred

Project Overview

A UK mid-market law firm with 180 fee earners spent an average of 3.2 hours per transaction on document review...

Technology Stack

PythonOllama (Llama 3.1 70B)Nomic Embed TextChroma DBLangChainPyMuPDFFastAPIReacton-premises Nvidia A100

Compliance & Standards

Legal Professional Privilege (LPP)UK GDPR (on-premises — no external processors)SRA Code of ConductCyber Essentials
Step 01

The Challenge

A UK mid-market law firm with 180 fee earners spent an average of 3.2 hours per transaction on document review — reading Land Registry title registers, mortgage offers, planning consents, and search results to identify relevant clauses, conditions, and risks. The senior partner estimated this represented £2.4M per year in fee earner time that could be better spent on client-facing work. Legal Professional Privilege (LPP) constraints meant client documents could not be sent to any external AI service without client consent. Budget: £120,000.

Step 02

Our Approach

LPP requirement

  • no client document data can leave the firm's infrastructure.
  • ClickMasters deployed a self-hosted

LLM stack

  • Ollama running Llama 3.1 70B on the firm's on-premises GPU server (Nvidia A100).
  • All inference runs locally — no data leaves the firm's network.

UK GDPR

on-premises processing eliminates the Article 28 DPA requirement for the AI model itself.

PDF ingestion

PyMuPDF for structured extraction, then chunking (512 token chunks with 50-token overlap for context continuity).

Embeddings

local Nomic Embed Text model (also on-premises).

Vector store

Chroma DB (self-hosted).

RAG pipeline

LangChain orchestrating retrieval from vector store + Llama 3.1 70B generation.

Output

  • structured JSON per document type.
  • LegalTech-

Specific Prompting

Document-type-specific prompt templates: Land Registry title register (identify charges, covenants, easements, restrictions), planning search (identify restrictions, conditions, planning history), mortgage offer (identify conditions precedent, terms).

Confidence scoring

each extracted clause includes a confidence score — low-confidence extractions flagged for fee earner review.

SRA Compliance and Auditability

Fee earner must review and approve all AI-generated summaries before use.

Audit trail

every AI summary linked to the source document, extraction timestamp, and approving fee earner.

SRA competency

AI is a tool assisting review, not replacing it — regulatory compliance maintained.

Step 03

The Results

Deployed at 12 weeks, £108,000 — under budget.

Document review time: 3.2 hours42 minutes average (87% reduction).

Fee earner capacity: equivalent of 4.2 additional fee earners without headcount.

LPP: zero client data left the firm's infrastructure.

SRA audit: commended for auditability of AI use.

Estimated annual value: £1.9M fee earner time released for client-facing work.

Client Testimonial
The on-premises requirement was non-negotiable for LPP compliance. Every other vendor wanted to send our client documents to a cloud API. ClickMasters built it entirely on our own infrastructure. The 87% reduction in review time is genuinely transformative for our practice economics." — Managing Partner, UK Mid-Market Law Firm (name withheld)
ClickMasters Case Study Team
Reviewed by James Whitmore, CTO

Project Details

Sector
LegalTech
Country
UK
Status
On Time
Contract
Fixed Price
Tech Stack
9 Technologies
Reading Time
6 min
IP Ownership
100% transferred
Last Updated
June 2025
Written By
ClickMasters Case Study Team
Reviewed By
James Whitmore, CTO

Related Case Studies

View All

Ready to Transform Your Business?

Let's discuss how our technical expertise can help you achieve remarkable results.