THE PROBLEM
$650B+ is being committed to AI infrastructure globally. Every regulated institution deploying AI faces the same unsolved problem: who independently verifies the AI works, who signs off, and who carries personal liability when it fails?
UK SM&CR makes Senior Managers personally and criminally liable for AI failures. The EU AI Act mandates conformity assessments by August 2026. Lloyd's is excluding GenAI from insurance policies entirely. Every bank, insurer, and regulated firm deploying AI needs independent verification — and no infrastructure-grade solution exists.
We're building it. The independent verification layer between AI systems and the regulated world. Not compliance SaaS. Not a dashboard. Not an LLM wrapper. Think what Moody's built for credit ratings, but for AI systems.
──────────────────────────────────────
THE ROLE
You'd be the technical co-founder. First engineer. Architect. Builder. You design the system, write the first line of code, ship the MVP, build the data flywheel, then hire and lead the engineering team as CTO.
The CEO has the regulatory network (King's Counsel, ex-FCA regulators, Lloyd's underwriters, senior compliance officers), warm client relationships, and a 33-version operating plan stress-tested to investor grade. What doesn't exist yet is the product. That's yours.
The division is clean: the CEO sells while you build. You own every technical decision.
──────────────────────────────────────
WHAT YOU'D BUILD
1. Verification Platform — Client-facing portal where regulated institutions submit AI systems for independent verification. Secure document upload, expert review workflow, branded verification reports.
2. Expert Annotation Workbench — The tool that captures structured judgments from domain experts (ex-regulators, barristers, actuaries) across multiple dimensions: binary judgments, graded assessments, reasoning chains, preference pairs, and regulatory mapping. This is the data engine.
3. RLHF / SFT Training Pipeline — The system that converts expert annotations into ML training data. Supervised fine-tuning on reasoning chains. Direct Preference Optimisation on expert preference pairs. Active learning that routes low-confidence cases back to humans. This is the technical moat.
4. Verification AI — The model that pre-screens AI systems before human review, handling routine verifications autonomously and routing ambiguous cases to experts. Target: 30% automation within 90 days, scaling to 60%+ within a year.
5. Sovereign Node Infrastructure — Jurisdictionally isolated cloud infrastructure with structural immunity to foreign jurisdictional overreach. Region-locked encryption, tamper-proof audit trails, per-client data isolation. Designed for UK first, then EU, GCC, and US expansion.
6. Multi-Agent System Telemetry — Real-time monitoring for fleets of AI agents interacting in production. Dependency graph construction, emergent behaviour detection, cascade kill switch capabilities. This is the Year 2-3 product that positions the company as national resilience infrastructure.
──────────────────────────────────────
THE HARD PROBLEMS YOU'D OWN
→ How do you verify AI systems that retrain weekly? Static audits break. You need continuous evaluation against drifting models.
→ How do you design annotation schemas that capture regulatory nuance across jurisdictions without collapsing expert knowledge into generic labels?
→ How do you build a data flywheel where every expert engagement generates training data that makes the next automated review more accurate — and prove that accuracy to regulators?
→ How do you architect sovereign infrastructure with structural immunity to foreign jurisdictional overreach while maintaining production-grade reliability?
→ How do you detect emergent behaviour when 100+ AI agents interact in production, creating outcomes no individual agent was designed to produce?
──────────────────────────────────────
WHO YOU ARE
We value ambition, shipping ability, and intellectual depth over pedigree. You don't need to come from a big AI lab. You need to have built real systems and want to build something that matters.
Must-haves:
→ Write production Python daily — systems that serve users, not notebooks
→ 4+ years building software systems, at least 2 involving ML/AI in production
→ Can design and deploy cloud infrastructure from scratch (AWS preferred)
→ Experience with LLM fine-tuning, model evaluation, or human-in-the-loop ML
→ Can architect a system on a whiteboard: data models, API design, trade-offs, failure modes
→ London-based or relocating. Full-time from Day 1
→ Comfortable with below-market cash in Year 1 in exchange for significant co-founder equity
Strong signals (not required):
→ RLHF, SFT, DPO, or reward model training experience
→ Regulated industry experience (financial services, insurance, healthcare, legal tech)
→ Previous CTO, co-founder, or founding engineer experience at a startup
→ Open-source contributions, published models, or shipped side projects
→ Experience designing for compliance — audit trails, data residency, encryption
You might be coming from:
→ An AI startup where you were CTO or senior engineer — especially one where you learned to ship fast with nothing
→ AISI or a frontier AI safety team where you've done evaluation/RLHF work and want to apply it commercially
→ A London fintech where you built ML systems in a regulated environment and want to go founding
→ A strong UK CS/ML programme where you've been building since undergrad
→ A compliance or security tech company where you understand the intersection of ML and regulation
→ An EF or Antler cohort where your first startup didn't work out but you're ready for your next founding role
──────────────────────────────────────
THE OFFER
→ Significant co-founder equity — genuine co-founder allocation, not an employee option grant
→ Below-market cash Year 1, scaling with revenue — we're honest about this
→ Co-Founder & CTO title with full technical ownership
→ In every investor meeting — you present the technical vision to Tier 1 VCs
→ You hire the engineering team as revenue grows
→ Bootstrap model: founders retain majority equity because we don't give it away early
──────────────────────────────────────
HOW TO APPLY
Skip the cover letter. Send:
(1) What you've built — be specific. A GitHub link, a product, a deployed system, a paper.
(2) Why this problem interests you — 2-3 sentences is fine.
(3) Your availability — when could you start full-time or part-time?