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
- Dataset
- Fine-tuning
- Evaluation
- Quantization
- 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 →
Code and models are on GitHub and Hugging Face.
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