Marc Haenle

Where technical leadership meets real-world AI application.

IT leader with 15+ years managing infrastructure, teams, and digital transformation. Working with AI/ML since 2018, formalized with an intensive Data Science program in 2025. Currently exploring roles where technical leadership meets data science application.

7/7 AI Skills in Production

15+ Years IT Leadership

14 Max Team Size

840h Data Science Training

The 7 AI Skills, Proven in Production

Based on analysis of hundreds of AI job postings (Anthropic, Scale AI, Upwork, Glean), mapped to my artifacts.

Specification Precision

Precise specifications for AI agents instead of vague requests. Not “improve customer support,” but: tier-1 tickets, escalation rules, sentiment scoring, reason codes.

Multi-tool system prompts for domain-specific research + agent specification frameworks

Evaluation & Quality Judgment

Error detection with fluency. AI is confidently wrong, and the skill is not mistaking fluency for correctness. Empirical evaluation against functional criteria, not gut feeling.

Systematic model evaluation (5 LLMs benchmarked), reranking optimization, automated eval harnesses

Task Decomposition & Multi-Agent Delegation

Breaking complex tasks into manageable segments and delegating to specialized agents. A management skill, but agents need exact specs, not vague briefings.

Two-tier security agent system, multi-stage RAG pipeline, autonomous knowledge maintenance agent

Failure Pattern Recognition

Knowing and diagnosing the 6 typical AI failure modes: Context Degradation, Specification Drift, Sycophantic Confirmation, Tool Selection Errors, Cascading & Silent Failure.

All 6 failure modes experienced, diagnosed, fixed, and documented as reusable lessons

Trust & Security Design

Where to deploy AI, where to keep humans? Graduated trust boundaries based on cost of error, reversibility, frequency, and functional verifiability.

Layered autonomous defense with graduated permissions, semantic vs. functional correctness verification

Context Architecture

The Dewey Decimal System for agents. Persistent vs. session context, clean data, searchable structures. The most valuable skill, because it enables not one system but dozens.

Million-scale vector search (pgvector/HNSW) + persistent agent memory with semantic search

Cost & Token Economics

Calculate ROI before you build. Model choice, blended cost, dynamic pricing. High school math, applied in an extremely fast-moving environment.

Hybrid ARM/x86 infrastructure tiering, local reranking (zero API cost), cost-driven database migrations

Career Arc

The 7 AI skills don’t emerge in a vacuum. They build on capabilities grown over 20+ years.

2001 – 2018

IT Administrator → IT Leader → Head of IT Service Management

Infopark AG · Home24 · Mister Spex

17 years of infrastructure, teams up to 14, data centers, DevOps, incident/problem/change management. The foundation for failure pattern recognition and trust design.

2019 – 2023

IT Operation Director

Messe Berlin GmbH

Budget ownership, strategic IT development, vendor management, data center restructuring. The foundation for cost & token economics.

2023 – 2025

Department Head, Infrastructure & Service

Kassenärztliche Vereinigung Berlin

Hybrid IT transformation, AI/ML implementation in a regulated environment (healthcare), 11 admins led with agile methodologies.

2025

Data Science Professional

Data Science Retreat, Berlin

840 hours intensive: Python, ML, Deep Learning, NLP, MLOps. The bridge from “understanding AI” to “building AI yourself”.

2025 – present

Freelance AI Infrastructure & Data Science

Freelancer, Berlin

Building production RAG pipelines, multi-agent systems, hybrid Kubernetes clusters, and AI-driven security solutions for clients across regulated industries. NDA-protected engagements.

Artifacts

Production systems built for clients in regulated industries. Project names and client details under NDA.

⚙

Enterprise RAG Pipeline

Domain-specific retrieval-augmented generation system. Million-scale vector search with pgvector/HNSW indexing, MMR diversity reranking, automated daily ingestion pipeline.

pgvector Embeddings LLM Tool-Use Python MLOps

⎈

Hybrid Kubernetes Infrastructure

Multi-node K3s cluster with mixed ARM64/x86_64 architecture. Cost-optimized workload placement, persistent storage, automated backup and disaster recovery.

K3s ARM64/x86 Infrastructure as Code Monitoring

⚖

AI Agent Memory & Integration

Custom MCP server infrastructure providing persistent agent memory with semantic search, knowledge graph integration, and plugin architecture. Token-efficient by design.

MCP Semantic Search Knowledge Graph Agent Architecture

⚠

AI-Driven Security Operations

Two-tier multi-agent system for autonomous threat detection. Real-time log enrichment with MITRE ATT&CK® mapping, cross-temporal attack chain correlation, automated incident response.

MITRE ATT&CK® SIEM Multi-Agent FastAPI

The Leadership Multiplier

The 7 skills are individual contributor skills. The multiplier is the ability to scale them across a team.

AI Skill	Grown From	Leadership Dimension
Specification Precision	Cross-departmental projects (Messe Berlin, KV Berlin)	Setting standards that keep an entire team consistent
Evaluation & Quality	SLA adherence, audit experience (Mister Spex, Home24)	Building review culture, eval frameworks as team assets
Task Decomposition	Teams up to 14, multi-provider mgmt, PRINCE2	Architecture decisions enabling team parallelism
Failure Patterns	17 years infrastructure ops, data center operations	Post-mortems that make the team immune
Trust & Security	Compliance (healthcare), change management	Policies that scale without slowing down
Context Architecture	ERP, data management, service catalog design	Infrastructure giving everyone on the team access
Cost & Token Economics	Budget planning, software audits (4.5 years Messe Berlin)	Budget ownership, ROI reporting to stakeholders