Dr.-Ing. Reinhard Munz

CTO @ Daiki | Distributed Systems & LLM Infrastructure | Go · Python · Kubernetes · AWS

Professional

I ship production code daily, increasingly by directing AI coding agents, with verification at the level of system behavior. As CTO of Daiki I lead development of a conformance-management platform — for AI systems under the EU AI Act as well as ISO certifications of regulated software such as Software as a Medical Device (SaMD) — helping companies identify their regulatory duties and document their systems in a form certifying bodies can consume: Daiki's SaaS solution. Python and Go backends, Kubernetes, Postgres, RAG pipelines on private LLMs.

Systems I have recently architected and built for Daiki clients: a redundant market-data and trading platform, written from scratch in Go — Kafka message bus, replicated QuestDB and Postgres, idempotent components under at-least-once delivery, institution-level FIX connectivity to exchanges and liquidity providers, and downstream AI-driven strategy, portfolio, and order-management components. An investor data-sharing platform on AWS — isolated per-entity EC2 environments with controlled sharing through a central routing instance, RDS Postgres, and Bedrock-powered AI and chatbot features. A security-hardened Kubernetes cluster for sensitive medical data — full-disk encryption at rest, all nodes hidden, with SSH reachable only via single-packet authorization (SPA).

Before Daiki, I worked at Gradient0 on copious other AI/ML projects. For further details, please see my CV.

Of course I have not been working alone. Kudos to the amiable and outstanding teams of both, Daiki and Gradient0!

Special Thanks go to Felix, Jona, Artur, Wolfgang, Kevin, Sandra, Julia, Paolo, Craig, Dominik, Scarlet, Carla, Dominic, Antoine, Morten, Karim, Denis, Clemens and everyone else!

Research

My research interests revolved around systems and how they can be improved to benefit users, providers, and the environment. I was working in the area of system security and privacy, with a particular focus on private data analytics. 

In my thesis I presented the first steps towards usable privacy protection mechanisms for data analytics systems. Usable in that context meant more queries and less distortion of answers than previous systems. I developed the UniTraX system, the first differentially private analytics system that is able to support personalized privacy budgets without giving up on answer accuracy.

With my thesis work I further laid the foundations for Diffix, a once commercially deployed private analytics system. Diffix used a defense in depth approach to privacy protection. It provided guarantees of usability to analysts and was one of the first systems to focus on usability in order to foster industry adoption of private analytics systems. Diffix later continued as Open Diffix.

Before my work on usable privacy with Paul, Deepak, and Matteo, I improved user satisfaction during peak system loads with Allen and devised energy efficient and failure resilient mechanisms for large-scale distributed environments with Umut and Dave.

During my time in different research groups I always had the pleasure to work with amazing people.

Some non-exhaustive sample: Arthur, Ekin, Ezgi, Fabienne, Imran, Iulian, Matt, Mike, Mustafa, Nancy, and Natacha.

Thanks folks! And thanks to all my other friends at MPI-SWS, CMU, and Uni Würzburg.

Publications

“Towards Usability in Private Data Analytics”. PhD thesis. Department of Computer Science, Technische Universität Kaiserslautern, 2019. [University publication server]

UniTraX: Protecting Data Privacy with Discoverable Biases. With Fabienne Eigner, Matteo Maffei, Paul Francis, and Deepak Garg. Conference on Principles of Security and Trust (POST), 2018. [Conference paper][Technical report with proofs][Gitlab repository]

Diffix: High-Utility Database Anonymization. With Paul Francis and Sebastian Probst Eide. Privacy Technologies and Policy (APF), 2017. [Conference paper][Preprint Diffix-Birch: Extending Diffix-Aspen 2019]

Persistent, Protected and Cached: Building Blocks for Main Memory Data Stores. With Iulian Moraru, David G. Andersen, Michael Kaminsky, Nathan Binkert, Niraj Tolia, and Parthasarathy Ranganathan. Carnegie Mellon University Parallel Data Lab Technical Report CMU-PDL-11-114v2, Nov 2012. [v2 Nov 2012][v1 Dec 2011]