QuantPi AI Validation Platform (PiCrystal)

QuantPi GmbH is a Saarbrücken, Germany company founded in 2020 by Philipp Adamidis, Dr. Antoine Gautier and Artur Suleymanov as a spin-off of the CISPA Helmholtz Center for Information Security and Saarland University after more than eight years of research. It provides a validation layer for enterprise AI that produces statistical evidence for release, change, compliance and certification decisions.

The platform, built around the PiCrystal testing engine, runs system-level and component tests on object detection, RAG, agentic, generative text, tabular and multimodal AI. It assesses data sufficiency with confidence intervals, uses calibrated AI-assisted annotation and what-if scenarios, applies a pre-built metrics library with judge-the-judge calibration, and outputs dashboards, model leaderboards, model cards and audit-ready reports. Customers include Deutsche Telekom, StepStone, Konux and BearingPoint.

• Multi-system testing: validates object detection, RAG, agents, generative text, tabular and multimodal AI.
• Statistical evidence: confidence intervals on results and on test data sufficiency.
• Scenario generation: AI-assisted annotation and what-if scenarios to widen test coverage.
• Judge calibration: benchmarks AI judges against a small golden dataset and factors in their error rate.
• Certification support: standard-aligned test scenarios and PDF or JSON reports with audit traces.
• Integrations: MCP server and API for agent workflows, CI/CD quality gates and GRC tools.

Vietnamese electronics and machinery exporters can validate vision-based inspection or robot perception models before EU conformity assessment. Banks can test RAG assistants and credit models for subgroup failures. HR platforms can audit recruitment AI for fairness under high-risk AI rules.

Results come with confidence intervals and audit traces that QuantPi states are pre-validated with certification bodies and market surveillance authorities, which shortens conformity assessment. Stress tests on blur, noise, weather and degraded inputs reveal failure modes hidden by a good global average. Self-hosted and air-gapped deployment keeps models and data inside the customer perimeter.

The EU AI Act and the EU Machinery Regulation make documented AI safety evidence mandatory for many products, and testing labs are being integrated into certification schemes for physical AI and autonomous machines. Southeast Asian exporters of electronics, vehicles and machinery to Europe will need comparable evidence for embedded AI.

QuantPi GmbH is registered at Halbergstr. 4, Saarbrücken (HRB 106305) and is EU-funded. It describes itself as TÜV SÜD's first CARAT partner laboratory worldwide, working under an ISO/IEC 17025-aligned quality system, and a member of the NVIDIA Halos AI Systems Inspection Lab. Partners include NVIDIA, HPE, CISPA, DEKRA and SGS TÜV Saar.