Table of contents:
Expert Panel Tackles AI Implementation
Demystifying Legal AI Architecture
The Hybrid System Approach
Building Safety Into AI Systems
Democratising Legal Services Through Technology
Regulatory Evolution and Innovation
Regional Innovation Ecosystems
Managing AI Risks in Legal Practice
Hallucination Prevention
Human Oversight Evolution
Iterative Improvement
Practical Adoption Guidance
Building the Future of Legal Services
Looking Ahead
Garfield AI CTO Reveals Technical Architecture Behind Legal AI at SRA Innovate Birmingham
Daniel Long shared deep technical insights on building the world's first SRA-regulated AI litigation platform, discussing hybrid system design, safety mechanisms, and the future of AI in legal services at the SRA Innovate event in Birmingham.
5 min
SRA Innovate Birmingham: Technical Deep Dive into Building Safe, Regulated Legal AI
Birmingham, 3 July 2025 – Daniel Long, CTO and co-founder of Garfield AI, provided technical insights into building regulated legal AI systems during a panel discussion at the SRA Innovate event in Birmingham. The discussion, led by Aileen Armstrong, Executive Director of Strategy, Innovation and External Affairs at the SRA, explored the practical realities of developing and deploying AI in the legal sector.
Expert Panel Tackles AI Implementation
The distinguished panel brought together key voices in legal innovation:
- Daniel Long - CTO and co-founder of Garfield AI, architect of the world's first SRA-regulated AI litigation platform
- Aisling O'Connell - Head of Innovation Policy at the SRA, shaping regulatory approaches to legal technology
- Clare Streets - Innovation and Digital Lead at SuperTech West Midlands, fostering regional tech collaboration
- Aileen Armstrong - Executive Director, Strategy, Innovation and External Affairs at the SRA (panel moderator)
Demystifying Legal AI Architecture
Dan provided a technical perspective on how Garfield AI achieved regulatory approval while maintaining innovation.
The Hybrid System Approach
"We describe our system as an expert system tied with an augmentation of large language models", Dan explained, challenging misconceptions about AI in legal services. Rather than relying solely on generative AI, the platform combines traditional software engineering with strategic LLM integration.
The architecture involves 20-30 different API calls working behind the scenes, each handling specific stages of the legal process. "We've built it on traditional software engineering principles with extensive unit testing", Dan emphasised, highlighting how this foundation ensures reliability and regulatory compliance.
Building Safety Into AI Systems
Dan detailed the multi-layered safety mechanisms that prevent AI errors from affecting legal outcomes:
Information Extraction Framework "We convert unstructured user input into structured data fitting defined schemas", Dan explained. This approach ensures that all information is validated before processing, preventing hallucinations from entering the system.
Logic-Based Validation Traditional software verifies fundamental legal requirements—limitation periods, claim amounts, court jurisdictions—before any AI-generated content is produced. "The expert system ensures LLMs cannot make inappropriate decisions or break court rules", Dan noted.
Continuous Testing Architecture The team developed innovative testing approaches, including "AI testing" that simulates unpredictable user behavior. "We run multiple parallel conversations several times daily to spot issues", Dan revealed, describing how the system proactively identifies potential problems.
Democratising Legal Services Through Technology
The panel explored how AI is making previously uneconomical legal work viable, with Dan emphasising Garfield's focused approach: "We deliberately limited ourselves to small claims debt recovery to maintain quality. It's a closed problem space with clear rules and procedures."
This strategic constraint enables the platform to deliver consistent, high-quality outcomes while operating within regulatory requirements. "People will always want things that are faster and smoother", Dan observed, noting how technology can transform access to justice.
Regulatory Evolution and Innovation
Aisling O'Connell provided crucial context on the SRA's approach to AI regulation, maintaining that the regulator remains technology-neutral while ensuring consumer protection. "We regulate professionals, not tools", O'Connell explained, "but the responsibility for outputs remains with regulated firms, regardless of the technology used."
This flexible approach has enabled innovations like Garfield AI while maintaining professional standards. The eight-month authorisation process Garfield underwent demonstrates the SRA's thorough yet supportive approach to novel technologies.
Regional Innovation Ecosystems
Clare Streets highlighted how initiatives like SuperTech West Midlands create safe spaces for legal innovation. "Small firms can be more agile in adopting new technologies compared to larger organisations, Streets noted, emphasising the importance of peer-to-peer learning and regional collaboration.
The discussion revealed how regional tech hubs provide crucial support for legal innovators, offering resources, connections, and validation that accelerate responsible innovation.
Managing AI Risks in Legal Practice
Dan shared candid insights about risk management in AI systems:
Hallucination Prevention
"Multiple stages of processing prevent single-point AI errors", Dan explained. Each stage validates and cross-checks information, ensuring no single AI decision can compromise legal accuracy.
Human Oversight Evolution
Currently, all Garfield outputs are reviewed by Philip Young, the company's qualified solicitor co-founder. "We're planning to move to a sampling-based approach as we scale", Dan revealed, demonstrating how human oversight can evolve with system maturity.
Iterative Improvement
The system design allows for continuous refinement based on real-world usage. "Every interaction teaches us something new about edge cases and user needs", Dan noted.
Practical Adoption Guidance
The panel provided actionable advice for firms considering AI adoption:
- Start with Clear Problems - Focus on specific, well-defined challenges rather than broad transformation
- Understand Your Constraints - Work within regulatory requirements from day one
- Build Incrementally - Test and validate at each stage of development
- Maintain Human Oversight - Ensure appropriate supervision remains in place
- Learn from Others - Engage with peer networks and regional initiatives
Building the Future of Legal Services
The SRA Innovate Birmingham event demonstrated the legal sector's growing sophistication in approaching AI adoption. By bringing together technical innovators, regulators, and regional supporters, the discussion illuminated pathways for responsible innovation.
"We're not just building technology", Dan concluded, "we're reimagining how legal services can work in the 21st century while maintaining the profession's core values."
Looking Ahead
As AI continues to evolve, events like SRA Innovate provide crucial forums for sharing knowledge and shaping the future of legal services. The combination of technical expertise, regulatory insight, and practical experience showcased in Birmingham offers a roadmap for successful AI implementation in law.
For more information about Garfield AI's technical approach to legal automation, visit garfield.digital.
About Garfield AI
Garfield AI is the world's first SRA-regulated AI litigation platform, combining expert systems with large language models to automate small claims processes. The platform helps SMEs recover unpaid debts through innovative technology while maintaining the highest standards of legal practice.
Media Contact:
Daniel Long, CTO - dan@garfield.digital
Philip Young, CEO - philip@garfield.digital
About the Author
Hugo Rawling
Legal Engineer
