About Ralph Schwehr.

Over 25 years in digital business models. Platform scaling, eCommerce and growth, AI and data. Strategy and execution in one person.

I get called in when several stakeholders are involved, decisions need to happen quickly, and impact has to be measurable. Revenue, margin or EBIT, OPEX, time-to-market, risk or compliance.

My standard is simple: measurable business impact, with governance that enables innovation instead of blocking it.

Personally I'm a tech person, a family person, and happiest in the mountains. Professionally I stay close to the hotspots of innovation. I completed the Data Science Program at HarvardX (Harvard University). My network reaches into Silicon Valley.

Education

Two stations, one foundation.

Business informatics as the architecture. Data science as the discipline.

  • Harvard University, HarvardXData Science ProgramCambridge, Massachusetts
  • AKAD UniversityM.Sc. (Diplom), Business InformaticsStuttgart

What I work on.

AI systems must be measurable. Evaluation before scaling.

from the decision down to operations

Strategy and leadership

  • AI strategy and operating model
  • AI governance: EU AI Act, GDPR, NIS2
  • P&L ownership and business cases
  • New business from service and data
  • Technology due diligence
  • Change leadership up to board level
Framework
EU AI Act, GDPR, NIS2, EU-sovereign operations

Models

Frontier, in the cloud
GPT, Claude, Gemini
Open, local
Llama, Mistral, Qwen, GLM
Own
oakai:alpha10, oakai:alpha11

Fine-tuned open models for German-language business applications, on Hugging Face.

Model-agnostic: I choose models by my own measurements of latency, cost, accuracy and safety. Prompts stay portable.

LLM engineering

  • LLM architecture: RAG, agents, buy or build
  • Tool calling and structured outputs with validation
  • Orchestration of multiple models with local fallback
  • Robustness tests for prompts
  • Evaluation for consistency and bias, blind tests with human reference
  • Agentic software development
  • LLMOps
Model pipeline
  1. Dataset
  2. Fine-tuning
  3. Evaluation
  4. Quantization
  5. Operation locally and in the cloud
Tools
Hugging Face, LoRA and QLoRA, MLX, GGUF, Ollama

Data and decisions

  • Simulation and decisions under uncertainty
  • Causal impact and experiments
  • Evaluation design
  • Statistical modeling
  • KPI systems that expose measurement errors
  • Data products traceable to the source
  • Statistics and machine learning
Tools
R, Python, SQL

Software and platform

  • Digital platforms and commerce
  • Multi-tenant SaaS with Stripe
  • Data platforms and API integration
  • EU-sovereign operations with PII masking
  • Security by design
  • Cloud and own hardware
Frontend
Next.js, React, TypeScript, Streamlit
Backend
FastAPI, Flask, PostgreSQL, Firestore
Cloud and operations
GCP (Cloud Run, Cloud SQL), AWS, Azure, Docker, GitHub Actions, Cloudflare

An example of simulation and decision models: the KORE simulator for my book Minimal Seed (working title) compares three strategies across 10,000 simulated journeys and shows how each result is calculated. Open the mission simulation →

Leadership

  • Responsibility through transitions and crises
  • Turnaround and restructuring
  • Team building and enablement
  • KPI and OKR management
  • Coaching and sparring
  • Negotiation

Business impact

  • eCommerce and marketplace scaling
  • Growth through acquisition and retention
  • CRM and lifecycle
  • Pricing and packaging
  • Process and OPEX optimisation
  • PMI and carve-out