AI that knows your business.
Governed knowledge assistants, private AI and business workflows that combine deterministic logic, AI capabilities and human judgment. From architecture and model evaluation to controlled deployment and operations.
Eight capabilities, from first use case to governed operations.
Combine the capabilities your project needs. Each has a defined role, controls and deliverables.
Knowledge AI & RAG
Indexed internal knowledge and approved external retrieval, with rights, provenance and grounding checks.
Explore this capability02
Sovereign & Private AI
Choose model, deployment and dependencies around data residency, control, licensing and portability.
Explore this capability03
Business Workflow Orchestration
Deterministic steps, AI proposals and human approval gates in one controlled business process.
Explore this capability04Deterministic Automation
n8n, Make or code for repeatable integrations, calculations and business rules.
Explore this capability05
AI Governance & Assurance
Define intended use, risks, evaluation criteria, responsibilities and evidence for the applicable review.
Explore this capability06Agent-ready Software
MCP servers and scoped APIs that expose existing products to authorised tools and agents.
Explore this capability07
Model Evaluation & Fine-tuning
Compare models against a documented task protocol; justify adaptation with measured results.
Explore this capability08
AI Infrastructure & Operations
Compute, storage, AI serving and monitoring on private, cloud or hybrid infrastructure.
Explore this capabilityThe model proposes; execution permissions are enforced outside the model. Sensitive actions require an authorised human decision bound to the action and its parameters.
Knowledge, action and assurance. Connected.
The visual explains the role of each capability. The detailed pages show the control branches, review decisions and known limits.
Governed knowledge
Internal authority and external evidence remain distinguishable. Retrieval checks rights, freshness and provenance before context reaches the model.
Inspect the architectureControlled business action
Deterministic rules bound AI proposals. Risk-based routing selects approved automations or a human checkpoint before sensitive execution.
Inspect the architecture
AI assurance
Evaluate the intended use, document limitations and residual risks, and prepare evidence for release and the applicable assessment route.
Inspect the architectureGenerative AI is everywhere. Real results are not.
Most companies run into the same three problems.
“Public AI doesn’t know our processes.”
A general-purpose model may lack the evidence needed for your procedures, contracts or case files. Retrieval coverage and source quality need to be evaluated.
“We can’t put our data in it.”
Sensitive information needs a defined processing scope: access rights, location, provider terms, egress rules and accountable owners.
“We lose hours on repetitive tasks.”
Manual reports and disconnected tools consume time. The solution needs repeatable rules, error handling and review gates, as well as AI where it adds value.
Three building blocks. One partner.
Deploy one service on its own or combine all three, from strategy to production.
Your knowledge. A governed view of the world.
We combine indexed internal knowledge with approved external sources when the use case requires it. Access rights, freshness and provenance shape retrieval; grounding checks distinguish supporting evidence from model inference.
- Hybrid lexical and vector retrieval, selected for the corpus
- Access-aware retrieval and versioned source metadata
- Optional governed external retrieval, with approved sources and controlled queries
- Evidence checks, conflict handling, abstention and critical-use review
Conceptual illustration. Inspect the detailed architecture and its limits on the service page.
Private AI, with its boundaries defined.
Choose open-weight or proprietary models and an on-premise, cloud or hybrid architecture according to the use case. Residency, administration, keys, licensing and dependencies are reviewed explicitly.
- Deployment options assessed against your processing policy
- Model and data versions tracked through evaluation
- Explicit provider terms and permitted external connections
- Operations, portability and upgrade qualification
Conceptual illustration. Inspect the detailed architecture and its limits on the service page.
Rules, AI and people. Working as one.
We design business workflows that combine deterministic logic, AI capabilities and human review. Evidence checks, permissions and approval gates define what can run automatically and what needs a decision.
- Deterministic algorithms and repeatable workflow steps
- AI proposals with evidence and policy checks
- Approval bound to the exact sensitive action and parameters
- Scoped execution, audit trails, incident handling and recovery where feasible
Conceptual illustration. Inspect the detailed architecture and its limits on the service page.
Make your software agent-ready.
Your product will be used by AI agents before your competitors notice. We expose existing platforms to AI agents through MCP servers and agent-ready APIs, with authentication, scopes, rate limits and usage metering. We did it for our own products, CLM Builder and PrezManager, before offering it to clients.
Choose the architecture for your constraints.
Managed services and private deployments each have trade-offs. We assess the actual provider, configuration and intended use rather than making blanket comparisons.
| Decision | Managed service | Private or dedicated architecture |
|---|---|---|
| Data processing | Review provider terms, configuration and location | Define hosting, access and permitted data flows |
| Knowledge and permissions | Assess available connectors and access enforcement | Design indexing and retrieval around your roles |
| Workflow actions | Assess tool permissions and approval features | Enforce action scopes and risk-based review |
| Quality and costs | Benchmark the chosen model and service plan | Benchmark the model, runtime and operating costs |
| Assurance | Determine evidence and responsibilities for the use | Prepare system-specific evaluation and release records |
Built for demanding industries
For over 15 years we have worked with healthcare and finance leaders. AI fits into environments we already know well.
& MedTech
- Regulatory & medical assistant with cited sources (SmPC, SOPs, studies)
- Pre-check of promotional content before MLR review
- Rep call preparation and CLM presentations
- Veeva ↔ IQVIA content conversion & migration
- Automated omnichannel reporting
& FinTech
- Instant search across internal policies and regulations
- Contract, KYC and client-file analysis & summaries
- Advisor support: sourced answers in seconds
- Automated fraud & security awareness
- Private hosting selected for your processing policy
We use it ourselves first.
Before offering it to clients, we automated our own processes. Some real examples:
Automated Jira project tracking
Every Wednesday and Friday at 9am, an n8n workflow compiles ticket progress for the whole team across all projects and emails each person a personalised summary.
LinkedIn × Google Analytics dashboard
LinkedIn metrics (followers, impressions, engagement) are combined with GA4 traffic and product-page views in a monthly report that’s ready to share.
Bridge AI: Veeva ↔ IQVIA converter
Developed by our healthcare company, Pixels Health, this proprietary tool uses AI to convert and migrate CLM presentations between the two leading pharma platforms.
Lead scoring and outreach agent
Apollo.io, LinkedIn Sales Navigator and Claude combined in n8n to score and route inbound and outbound leads.
How much time could you win back?
Move the sliders to match your situation. Indicative estimate.
From first call to production
A pragmatic approach: start small, prove the value, then scale.
Workshop to identify the highest-ROI use cases and your data constraints.
Architecture, hosting, data sources and success metrics.
A first assistant or workflow tested by your teams on real data.
Security hardening, integration with your tools, user training.
Monitoring, maintenance and new use cases as your priorities evolve.
Pick your starting point
Three packages to move at your own pace. Every quote is tailored to your scope.
Automate
Save time fast on repetitive tasks.
2–4 weeks · On quote- Audit of 5 to 10 processes
- 2 to 3 n8n or Make workflows in production
- Automated reporting (email / Teams)
- Documentation + 1 training session
- 1 month of support included
Knowledge AI
Give your teams an AI assistant that knows all your documentation.
4–8 weeks · On quote- RAG assistant on your document sources
- Cited answers and access-rights management
- Swiss / EU hosting
- Web interface or Microsoft Teams integration
- Quality evaluation + user training
- 3 months of support included
Sovereign AI
Your own private AI platform, across the organisation.
Tailored programme · On quote- Open-source models on your infrastructure
- Multiple RAG assistants by department
- Connected automations and AI agents
- Security, traceability, DPO support
- AI governance and roadmap
- Run & maintenance under SLA
AI Act Readiness
Know what you run and prove how it behaves.
2 weeks · Fixed price- AI system inventory
- Risk classification
- Transparency and logging gap analysis
- 12-month roadmap
- DPO briefing
The stack behind our AI, automation and apps.
Open-source and commercial models, automation platforms and the frameworks we have mastered since 2010. We pick the right tool for your data and your constraints, with no vendor bias.
Your questions, answered
What is RAG, in plain terms?
RAG retrieves relevant evidence from selected sources and supplies it as context to a model. It does not erase the model’s learned knowledge or guarantee correct answers. We evaluate coverage, source support, access rights and handling of uncertainty.
Will my data be used to train the AI?
Data use depends on the deployment and provider terms. We define the permitted processing and training policy, check contracts and configuration, and control external connections accordingly.
Do we need an in-house technical team?
No. We handle the architecture, deployment, integration and maintenance, so your teams simply use the tool. If you want, we can also train an internal champion.
Why n8n rather than Make (or the other way round)?
n8n can be self-hosted in Switzerland, which makes it ideal for sensitive data and high volumes. Make is very quick to set up for simpler needs. We recommend whichever fits your context, with no vendor bias.
How soon will we see results?
A first automation workflow can be live in about two weeks. A RAG assistant on a defined document scope usually takes four to eight weeks.
Is it compatible with pharma compliance (MLR)?
AI can assist preparation and pre-checks in a defined workflow. Authorised reviewers retain the required medical, legal and regulatory decisions; the system is evaluated for its intended use and keeps proportionate evidence and records.
How much does it cost?
Every project is quoted after the free audit, based on the number of use cases, data sources and hosting option. We prefer clear fixed project fees over per-seat licences.
What is an AI agent and how do you keep it under control?
An agent proposes tasks or tool actions using a model. Deterministic permissions are enforced outside the model, with human approval for sensitive actions, audit records and recovery designed for each system.
Are we concerned by the EU AI Act as a Swiss company?
Applicability depends on the system, its intended use, the parties involved and where it is placed on the market or used. We help map the relevant requirements and coordinate the assessment with your legal and compliance owners.
What is MCP?
The Model Context Protocol is an open standard that lets AI agents use tools and data sources in a controlled way. We build MCP servers for existing products.
Can you run the models on our own servers?
Yes. Depending on the model licence and resource needs, deployment can use your own servers, a selected cloud or a hybrid architecture. Residency, dependencies and permitted external retrieval are assessed explicitly.
Explore the engineering, then discuss your case.
The AI Architecture Lab exposes a real small neural network, synthetic retrieval examples, simulated approval gates and deployment memory estimates. Its assumptions and limits remain visible.
Let’s make AI work for you.
In a free one-hour audit, we’ll identify your top three use cases, the time you can save and the right architecture for your data.