README.md

# Solbach Datentechnologie GmbH

Research-grade data infrastructure for B2B sales work in the DACH region.

We turn flat company lists into readable sales cases — combining registry data, website context, classification and signal layers into per-account research a sales team can actually use, instead of a CSV they still have to qualify.

Most of our compute is recurring batch inference across millions of DACH entities: extraction, classification, scoring, re-scoring. That workload is structurally different once you own the inference layer, so we run our own — currently vLLM with quantized 30–50B-class models, batched and queue-pressured to keep utilization high. A single tuned RTX PRO 6000 handles ~1B tokens/day for our workloads, roughly 5× our prior external-API baseline; most of the gain came from pipeline shape, not the model swap.

This profile is where we track the inference side of that stack and share useful artifacts where it makes sense.

Long-form notes on methodology, data and how the whole thing comes together: [blog.solbachleads.de](https://blog.solbachleads.de)

---

Solbach Datentechnologie GmbH · Bad Honnef, Germany