ralph schwehr

Downloads

Short CV via magic link.

Drop your name and email. The link leads to the current CV as a PDF.

I review every request personally. You will usually receive your download link by email within 24 hours.

By submitting, you confirm that name and email may be stored to send you the download link.

Production-grade model

oakai:alpha11: AI language model (LLM) for support quality.

oakai:alpha11 is an AI: a language model, or LLM for short. It scores support chats automatically across five dimensions and returns the result as strict JSON, ready to process downstream. I trained it on real, anonymized support conversations; a documented traceability test returned zero hits, no single conversation can be traced back to a person. Fine-tuned on Qwen2.5-7B, open-sourced under Apache 2.0. The honest step up from alpha10: that one was a proof of concept on synthetic data, alpha11 runs production-grade on real data.

0.52
avg. deviation (MAE, lower is better)
100%
valid JSON
1.5×
faster than the base model
View on Hugging Face →
ollama pull RaSchwehr/oakai-alpha11
Whitepaper (PDF) →

Proof of Concept

oakai:alpha10 - my own AI language model (LLM).

oakai:alpha10 was my first proof of concept: a small language model (LLM), trained myself and fine-tuned on Qwen2.5-7B, running entirely on your own hardware instead of in the cloud. Open-sourced under Apache 2.0, built for retail and customer analytics.

90%
Accuracy
0.34s
per classification
25×
faster than a 3× larger model
View on Hugging Face →
ollama pull RaSchwehr/oakai-alpha10
Whitepaper (PDF) →