Domain expertise + technical capability
The combination law firms and pension schemes need but rarely find.
William Crump
Background
UK Pensions Lawyer, qualified 2016.
9 years pensions law practice: in-house counsel and private practice. Defined benefit schemes, PPF assessments, regulatory compliance, trustee advisory.
Currently Legal Counsel at Pension Protection Fund.
Previously: OpenAI contract work (legal research and evaluation, ended January 2026).
Why This Matters
Most AI vendors do not understand pensions law. Most pensions lawyers do not build infrastructure.
The gap between what AI promises and what legal work requires is not intelligence. It is domain expertise. It is understanding what an obligation is. It is knowing when "must" in legislation creates a duty and when "may" gives discretion.
Generic tech vendors cannot build verified legal infrastructure. They do not know the domain. They do not read legislation. They cannot verify outputs because they do not understand what correctness means in regulatory context.
What I Built
A year deep in the question: what does AI need to be trustworthy in legal work?
Answer: verified reference infrastructure. Structure extracted from legislation. Obligations mapped to source provisions. Temporal tracking. Human verification at every step.
Not "AI that solves everything". Infrastructure that makes AI verifiable.
Not a Generic Tech Vendor
What This Is Not
- A startup trying to scale fast and sell to everyone
- A consultancy offering training programmes and workshops
- A software vendor with a chatbot product
- Someone who thinks AI alone solves legal complexity
What This Is
A qualified lawyer who spent a year building the infrastructure AI needs. Deep regulatory knowledge and technical capability. The rare combination that can extract structure from legislation, verify correctness, and build tools that legal professionals can trust.
The Approach
Methodology before tools. Systems thinking before deployment. Verification before release.
No vendor hype. No "revolutionary AI". No overselling. Honest about what works, what does not, and what the work requires.
45Black Limited
Company Details
45Black Limited
Company No. 16860363
20 Wenlock Road, London, England, N1 7GU
Founded: 2024
Location: UK
What 45Black Does
Builds knowledge infrastructure that makes AI trustworthy in regulatory work. Currently: UK pensions law. Future: other legal domains where verified structure matters.
Not a product company. Not a consultancy. Infrastructure work for organisations willing to invest in methodology before tools.
Recognition and Partnerships
Neo4j Founders Award — Selected for graph database innovation in legal infrastructure
AWS Activate Programme — Application pending for cloud infrastructure credits
RequirementOne Partnership — Collaborating on requirements engineering and systems methodology
Advisory — Dr David McKee (placeholder — may be joining as technical advisor)
How We Work
Honesty over hype. We make conservative claims we can defend. If we don't know, we say so.
Methodology over magic. AI is just the tool that makes systems thinking visible. The value is in the thinking.
Domain expertise matters. Generic AI consulting won't work here. Pensions law requires deep regulatory knowledge.
Capability transfer, not dependency. We teach the methodology. It stays with the team when we leave.
Attention is finite. Everything we build should reduce cognitive load, not add to it.
Technology Stack
What We Work With
Languages: TypeScript, Python
Databases: Neo4j (knowledge graphs), PostgreSQL, SQLite
Frameworks: Next.js, FastAPI, Hono.js
AI: Claude (Anthropic), OpenAI embeddings
Infrastructure: Cloudflare (Pages, Workers, Tunnel), AWS (EC2, S3)
Why This Matters
The technology is unremarkable. Neo4j exists. Python exists. The hard part is not the stack—it is the domain expertise to extract structure, verify correctness, and build infrastructure that legal professionals can trust.
We are model-agnostic but Claude-first for reasoning tasks. We do not sell AI models. We build the reference layer that makes any AI verifiable.
Contact
If You Need This
Register interest. We will schedule a conversation about your implementation challenges. We will assess whether systems thinking and verified infrastructure fit your needs.
No sales pitch. No diagnostic call. A conversation about whether this approach makes sense for your situation.