We help companies integrate technology into their operations, and we build the open-source and enterprise tools we wish existed while doing it.
From first workshop to the moment your team runs it on their own.
Workshops with your team to understand the problem, check what's technically possible, and estimate cost and return.
We build a prototype that actually runs on your real data, not on a demo set. You can evaluate it before we commit to scale. Vision, language, or forecasting.
We ship it live with monitoring, alerts, and fallback plans. Runs on your servers, in our EU cloud, or worldwide on request.
Full handover of the system: architecture, decisions, runbooks, and the reasoning behind each. We stay available as long as you need.
AI without the marketing Austrian Staatspreis Consulting 2022, a sub auspiciis PhD in AI, and one of the first AI-assisted lectures in Austria. Backed by four shipped patents and production deployments running since 2009.
Recent examples we can share publicly:
bds421 built the AI inside Cybertrap's deception mesh. Our models detect intruders early, profile each attacker's skill and intent, and adapt the decoy environment in real time. Every intrusion feeds a temporal knowledge graph so the next trap is smarter than the last.
An in-house bds421 neural network, "Hawkeye". First shipped on waste-collection trucks, then scaled to sorting-plant conveyors, where it detects 32 interfering materials in real time and judges each load's purity. The camera hardware is commodity smartphones, cheap and fast to replace. Robot-arm control to pull valuables off the belt automatically is the next phase of the project.
Austrian conglomerate spanning retail, industry, and print & publishing. Their first production AI. Logistics optimisation and real-time data analysis for the retail chains Libro and Pagro. The system ingests sales, inventory, and logistics signals across the chains' store network and central warehouses. Built in 2018 under the MTB group era.
Two product lines, same engineering.
ρ (rho) is the open-source family, Apache-2.0 on GitHub.
σ (sigma) is the enterprise line: full systems supported by the team that built them, deployed on-prem, in eu-central-1, or worldwide on request.
Inspired by the pi/llm adapter, reimplemented in Go. Production-grade, Apache-2.0. The building blocks we use on every consulting project.
Full systems for regulated and mission-critical environments. Deployed on-prem, in eu-central-1, or worldwide on request. Supported by the team that built them.
Across major LLMs, hallucination rates range from 3% to 23% under optimal conditions, and in 80–96% of cases where they should admit uncertainty, they answer anyway. datAInsights is built differently: every answer runs through Tyla, datAInsights' own reasoning engine with a temporal-first data model, and is verified before it reaches you. Not as a target but by construction, hallucination drops to almost zero. The formal foundations trace back to Datalog research at the University of Oxford and TU Wien (paper); the temporal-first extension to Markus Nissl's PhD.
Different domains, same engineering. Vision, LLM, forecasting, active defense. Applied wherever the problem surfaces. One team, many roles: PhDs, engineers, software architects, UI/UX experts, programmers, art directors.
Austria's national prize for outstanding, export-capable consulting. Awarded annually by the Bundesministerium für Arbeit und Wirtschaft together with the WKÖ Fachverband UBIT. Presented at the 20th IT- und Beratertag, Hofburg Vienna, September 2022.