RAG & Hybrid Retrieval
We build hybrid retrieval infrastructure for enterprises: semantic, keyword and reranking, with provenance and contradiction detection. The repository is open source.
Proof
Beats the McKinsey benchmark on retrieval accuracy
Four specific things we do in AI.
We research them, benchmark them and ship them with paying clients.
The capabilities
Learn more →
Training & Fine-Tuning
We train and fine-tune local small language models on an organisation's own governed knowledge. Sovereign AI: efficient, private and economically defensible.
Proof
Small models on governed knowledge, not generic data
Agentic Systems, CLI & MCPs
We build agentic systems, command-line tools and MCP servers that let AI take governed, audited action across your stack — verifiable, not just capable.
Proof
Verified agents, Catalyst, Control8
AI Applied Research
We apply AI to solve complex problems for companies and research organisations — from regulatory compliance to scientific discovery — with published, benchmarked work behind it.
Proof
Published papers and real deployments
Production-grade from the start.
Every KVA product ships on a battle-tested stack designed for performance, security, and developer experience.
Framework
Next.js 16
Language
TypeScript
Database
PostgreSQL
ORM
Prisma
Styling
Tailwind CSS
Testing
Vitest
Deployment
Vercel
Containerization
Docker
GDPR-aware by design
Built around European data-protection principles from the start
EU AI Act readiness
Designed for the EU AI Act, ahead of the August 2026 applicability deadline
SOC 2 Type II–aligned
Security controls aligned with SOC 2 Type II requirements (not yet certified)
Open Source First
Core tools published on GitHub. Transparency builds trust