What I build.
Three active projects at the intersection of AI, automation and domain expertise.
Negotiation Agents
Flight simulator for negotiators — AI-powered diplomatic simulations without financial or reputational risk.
The platform closes a gap in professional development: traditional negotiation training is expensive, time-intensive and creates no real counter-pressure. Negotiation Agents democratises access to sophisticated negotiation practice.
Core Features
- 41 behavioural archetypes based on Hofstede, Big Five and Harvard PON
- Dynamic topic sequencing — easier wins first, then escalation
- Red-line enforcement: AI agents cannot be persuaded beyond their limits
- Multi-channel: Signal, web, email or terminal — human and AI players mixed
- Complete audit trail with real-time fact-checking
PolicyMonitor Studio
EU policy monitoring and briefing platform — policy intelligence with evidence, not gut feeling.
PolicyMonitor automates the time-consuming process of EU source monitoring and transforms raw information into stakeholder-ready briefings with full source traceability. Currently in early access.
Core Features
- Continuous monitoring of selected EU sources with versioned history
- Evidence-based triage: updates pre-structured with context and source links
- AI search with citations: domain questions return answers with direct document references
- Report workflows: collection, templates, draft generation and export
- Governance: tenant isolation, audit trails and scoped API access
ShellAgents
AI-driven universal agent pipeline — entirely in Bash.
ShellAgents automates workflows with modular, agent-based shell scripts. LLMs dynamically plan team structure and workflow, shell scripts execute. Every action, every piece of data, every decision is file-based and auditable.
Architecture
- LLM generates team structure and workflow (laufzettel.json) from user request
- Shell scripts orchestrate execution and instantiate agents from archetypes
- All interactions, inputs and outputs in structured directories
- macOS / POSIX-compatible, Bash as primary scripting language
- Sequential execution — auditable and traceable by design
Typical Work Patterns
Anonymised patterns from consulting projects.
Operations Stabilisation
Problem: Ticket backlog, no prioritisation, no maintenance concept.
Result: Transparent operating model with clear escalation paths.
Security Resilience
Problem: Uncertainty about risks, backups never tested, no incident plan.
Result: Risk matrix, tested recovery, 90-day plan.
Architecture Clarity
Problem: Grown tool landscape, no big picture, restructuring feels risky.
Result: Target architecture, migration sequence, decision brief.