Cases

What we build and who we advise — practice, not PowerPoint.

Own developments

Our own AI systems as proof: we know what we're talking about.

Mingly — Multi-LLM Desktop App

Mingly brings Claude, GPT, and locally running Ollama models into a native macOS desktop app. Requests switch between cloud and local models without context loss — history is model-agnostic, sensitive content can stay on-device. In production since early 2026, written in TypeScript on Electron, stored in local SQLite.

ElectronReact 19TypeScriptTailwind

DocMind — RAG Frontend for Private Data

DocMind makes private document collections — PDFs, notes, internal wikis, contract archives — semantically searchable, fully offline. Embeddings are computed locally, Qdrant runs as a sidecar process inside the Electron app. User data does not leave the computer. Built for lawyers, fiduciaries, research teams, and SME leadership who need AI on confidential material without sending it to the cloud.

ElectronReactTypeScriptQdrant (Sidecar)

RAG-Wissen — Production RAG for our own knowledge base

RAG-Wissen is our internal RAG system for advisory knowledge, technical documentation, and market intelligence. Runs in production with Qdrant v1.17.1 and Harrier embeddings, provides the research base for our blog articles, and connects via an MCP server to Claude Code, Mingly, and our internal tools. Single source of truth for every advisory sparring session.

Python 3.13Qdrant 1.17.1FastAPIMCP

Prüfstand — AI Testing Framework

Prüfstand is our framework for systematic testing of AI systems: prompt variations, model comparisons, quality scoring via LLM-as-a-Judge, regression detection over time. Used for QA of our own products and in advisory work for customers who want to harden their own AI pipelines before going to production.

Electron GUIPython BackendVitestPytest

Nexbid — Open Infrastructure for Agentic Commerce

Nexbid is a two-sided marketplace for the moment AI agents buy: publishers bring content, advertisers bring products and campaigns, Nexbid connects both sides through open protocols (AdCP, MCP) instead of a walled garden. Live since April 2026 with discovery API, MCP server, 19 transaction tools, x402 payment rail, and wallet-pay magic-link. Founding publisher Betty Bossi, founding advertisers FOCUSWATER and The Champagne.

TypeScriptNode.jsVercel Functions (fra1)Neon Postgres

Compass — Research Project

Compass is our internal research project on agentic workflows for complex research tasks: translate multi-step work that a human would spread over days into an orchestrated agent flow — with control checkpoints, source provenance, and auditable reasoning. Written in Python on the Claude Agent SDK, with MCP-based tool connections to our internal knowledge sources.

Python 3.13Claude Agent SDKMCPPydantic

Eval-Framework — LLM-as-a-Judge with bias correction

Our framework for systematic evaluation of LLM outputs: pairwise comparisons, bias corrections (position swap, verbosity, self-preference), calibration against human baselines with Spearman ρ and Krippendorff α as acceptance thresholds. Used for RAG-Wissen quality gates, Nexbid content reviews, and advisory mandates where customers need objective AI quality measurement.

Python 3.13SQLitePydanticPrometheus2-Judge

Mineralis — AI-native Equity Research for Mining Equities

Mineralis is a SaaS platform for investment research in the global mining and energy market: every listed mining company in the world gets a self-updating AI research profile with source attribution, material events are detected and classified in real time, and an explainable scoring framework (MAS-Score) replaces black-box DCF. Positioned as 'Bloomberg for junior miners' — the depth and coverage of CapIQ/S&P (USD 30K/year) with AI-first architecture and transparent pricing.

Next.js 16 (App Router)Clerk v7Neon Postgres (aws-eu-central-1)Qdrant Cloud (eu-central-1)