Services
All products at a glance
One evaluation system, two service lines. Each engagement starts with a scoping call and ends with a reproducible report.
Not sure where to start?
Or take our 2-minute readiness check for a personalised recommendation.
Helvetic AI: Compliance
Regulatory AI Compliance
For Chief Risk Officers, Chief Compliance Officers, and compliance teams.
Entry
AI Risk Classification
EU AI Act risk classification with Swiss regulatory overlay (nDPA + FINMA). Structured assessment that determines risk level, applicable articles, compliance obligations, and timeline.
Recommended
Core
EU AI Act Technical Compliance Report
29 EU AI Act compliance benchmarks × 6 principles × 18 technical requirements. Per-article scores (Art. 9–15), traffic-light dashboard, gap identification, remediation priorities. Includes frontier model baselines for comparison.
Premium
FINMA Model Validation
Purpose-built evaluation against FINMA Guidance 08/2024. 5 categories (governance, operational risk, outsourcing, data quality, explainability), comprehensive FINMA-specific evaluation. Board-ready executive summary with risk heat map and remediation roadmap.
Helvetic AI: Performance
AI Model Evaluation
For CTOs, Heads of AI, and engineering teams.
Entry
AI Model Evaluation Report
5-model comparison: accuracy, Swiss language quality (DE/FR/IT), domain-specific scenarios, failure mode detection. Standard mode uses quarterly benchmark intelligence. Custom mode runs the evaluation against your model. You receive model rankings, head-to-head comparisons, and selection recommendation.
Comprehensive
Full SOTA Model Sweep
30+ models evaluated against Swiss-Bench + EU AI Act compliance suite + your domain. OpenAI, Anthropic, Google, Mistral, local models, Swiss models. Full ranking table, Swiss language quality, EU AI Act scores, total cost of ownership. The definitive comparison.
Add-ons
Add-on
Local AI Setup Advisor
Want to run AI models locally instead of relying on cloud APIs? We assess your use cases, recommend the right hardware and software stack, and deliver a complete deployment guide. Includes model selection per use case, a 3-year total cost of ownership comparison (local vs. cloud), and a security checklist for on-premise AI.
Add-on
Helvetic AI Select
We tested 50+ fine-tuned open-source models and selected four that beat their base models by 6–20 percentage points on domain benchmarks. You receive a model recommendation from our verified library, an independent benchmark report with base model comparison, Swiss language evaluation (DE/FR/IT), EU AI Act compliance assessment, and a deployment guide (GPU sizing, quantization, API setup).
- Cybersecurity, Finance, Medical domains available
- Models run locally. No data leaves your premises
- Custom fine-tuning on your data available on request
Implementation
From evaluation to execution.
You’ve identified the best model for each task. But how do you ensure every request is automatically handled by the right model, without manual intervention? The AI Model Router turns your evaluation results into executable routing rules, with compliance guardrails built in.
Entry
AI Model Router — Config
JSON/YAML routing rules derived from your evaluation results. Integration guide for your existing infrastructure. Route tasks to the right model based on language, domain, risk level, and cost constraints.
Recommended
Standard
AI Model Router — SDK
Python package with routing logic, audit logging, and automatic fallback. Integrates with your existing LLM calls. Every routing decision is logged for compliance and cost tracking.
Premium
AI Model Router — Proxy
Docker container, OpenAI-compatible API proxy. Drop-in replacement. Your applications don’t change. Automatic model selection, load balancing, cost optimization, and compliance enforcement.
Models change. Routing rules update automatically with each quarterly monitoring cycle →
Recurring
AI Model Monitoring
Quarterly re-evaluation of your production models. Available after initial compliance or performance engagement.
Recurring
AI Model Monitoring
Quarterly re-run: Swiss-Bench + compliance suite + domain-specific evals. Alerts for significant changes. Standard mode detects when model providers update capabilities.
Compliance Tracking
Score changes, regulatory posture shifts, new requirement mapping.
Performance Monitoring
Drift detection, new model releases, cost-performance ratio changes.
Resources
Free. Because transparency builds trust.
Quarterly Benchmark
Swiss-Bench Leaderboard
395 proprietary evaluation scenarios. DE/FR/IT. How do frontier models perform on Swiss-specific tasks? Updated quarterly.
View results →
Free Download
Quarterly Compliance Report
EU AI Act compliance suite + Swiss-Bench against 10–15 frontier models. EU AI Act compliance scores, Swiss language performance, FINMA alignment.
Request report →
Instant Score
Compliance Scorecard
Select a commercial model or submit an endpoint. Receive an instant compliance score. Overall score, 6-principle breakdown, traffic light rating.
Request scorecard →
These resources demonstrate our methodology. The same system produces your individual evaluation.
Data Handoff
5 delivery tiers.
| Tier | Mode | How It Works | For Whom |
|---|---|---|---|
| Intelligence | Benchmark Intelligence DEFAULT | We benchmark commercial models with our own keys. You receive scores + interpretation. No access needed. | Default for everyone |
| Standard | API Key | You provide endpoint for custom/self-hosted models. System runs on our infrastructure. | Custom model evaluation |
| Regulated | Docker | Dockerized system runs on your infrastructure. You send output logs. | Banks, insurers (FINMA) |
| Premium | On-Premise | We deploy hardware on your site. Full air-gap. | High-security, classified |
| Privacy-First | Anonymize-First | You run anonymization script (we provide). Send anonymized data. We evaluate. | Data-sensitive clients |
| Custom | Tailored to your infrastructure and compliance requirements | On request | |
Contact
contact@ai-helvetic.ch
Ready for an independent evaluation?
Start with an AI Risk Classification or a full AI Model Evaluation. Within one to two weeks you'll know where your AI systems stand: evidence-based, not opinion-based.