Florian Dietz - Scholar

Recommended expert

Saarbrücken, Germany

Experience

Jan 2025 - Mar 2025

3 months

San Francisco, United States

Scholar

MATS

Designed an automated model evaluation pipeline enabling LLMs to inspect each other for alignment issues
Implemented a RAG system with iterative cross-examination for reliable results
Automated generation of written summaries and hypotheses to support rapid iteration and hypothesis testing

Mar 2022 - Jun 2022

4 months

Senior Data Science Consultant

PwC (PricewaterhouseCoopers)

Helped unify and improve multiple ERP systems by extending development and adding functionalities
Advised employees on best practices for coding standards and recommended technologies to improve IT infrastructure efficiency
Integrated programs with a US-managed umbrella project, balancing different requirements and tech stacks

Jul 2021 - Present

4 years 7 months

PhD Student, Artificial Intelligence

Saarland University

Pursuing funded research on neural network capabilities for interpretability and generalization
Enhanced a foundation model with a novel arithmetic module enabling reliable math solving without external tools (outperformed larger models)
Created an analysis and debugging tool for neural networks, spotlighted at ICML 2024 Workshop for Mechanistic Interpretability
Developed an architectural variant improving compositional generalization, demonstrating evidence of grokking where Transformers failed
Research hypothesis focused on experimental thinking and iterative self-correction in AI systems

Dec 2020 - Sep 2021

10 months

Data Engineer & Full Stack Developer

AbbVie

Planned and implemented a data pipeline to merge and clean three separate published databases into one, enabling effective cross-department communication
Optimized bottleneck SQL queries by several orders of magnitude, allowing real-time data updates
Built a Django frontend to organize and display data, facilitating analysis and reducing redundancy
Proposed and implemented an automated tool to find and visualize connections in legacy databases
Contract extended multiple times due to continuous identification of new data utilization opportunities

Sep 2019 - Dec 2020

1 year 4 months

San Francisco, United States

Chief Technology Advisor

AI-bees

Advised on technology and hiring decisions and formulated long-term AI strategy for the startup

Sep 2019 - May 2020

9 months

Munich, Germany

AI Research Consultant

Huawei

Proposed and developed a tool to improve machine learning model synergy with simulation software and robustness against unexpected situations
Discovered anomalies in client data, proposed and implemented model modifications to extend capabilities
Contract extended despite the economic crisis caused by the COVID-19 pandemic

Mar 2017 - Sep 2019

2 years 7 months

Munich, Germany

Founder, CTO

Elody

Built an AI startup that automatically finds and executes appropriate software for user problems in the browser
Led customer and investor communications, cofounder vetting, and technology development
Managed the entire tech stack and trained developers on the platform
Focused initial offering on data science applications
Marketplace failed to reach critical user traction

Dec 2016 - Mar 2017

4 months

London, United Kingdom

Senior Machine Learning Engineer

Palantir

Worked on high-security data science projects under NDAs for prestigious clients (details confidential)

May 2015 - Nov 2016

1 year 7 months

Munich, Germany

Data Scientist

Volkswagen Data Lab

Developed novel AI algorithms optimizing for revenue-driving metrics rather than traditional ML metrics
Contributed to saving multiple million euros through custom solutions
Extracted and analyzed data from Oracle and PostgreSQL databases to solve stakeholder problems
Conceived and implemented end-to-end data analysis solutions

Jan 2009 - Present

17 years 1 month

Independent AI Researcher

Independent Research

Conducted AI and machine learning projects in spare time aiming for safe AGI
Founded Elody based on semi-automated programming methods
Advised companies on AI strategies
Explored parallels between human cognitive biases and algorithm failure modes

Senior Data Scientist

AE Studio

Worked as Data Scientist for client projects and AI researcher on internal AI alignment initiatives

Mentor

SPAR

Mentored two teams on alignment-related projects:
Split Personality Training: Built a second personality in an LLM to review past behavior with access to hidden internal knowledge
Deliberative Credit Assignment: Improved reasoning performance by making chain-of-thought interpretable, aligning evolutionary and financial incentives

Summary

AI Research Engineer with 15+ years of experience, bridging cutting-edge research and business value. Specialized in LLM architectures and neural network optimization.

AI TRANSFORMATION & STRATEGY: Consulted multiple companies on AI strategy, data science, and finding synergies between their existing tech stack and emerging technologies.
LLM EXPERTISE: Recent intensive upskilling on LLM-based workflows with APIs under the mentorship of industry leaders. Results included RAG, cross-examination between multiple LLMs, and various tricks that are little known outside of frontier labs.
DEEP UNDERSTANDING OF NEURAL NETWORKS: Fundamental research funded by the NHR. Created a popular tool for debugging neural networks, making them more interpretable.
BUSINESS KNOWLEDGE: Startup founder, freelance data science consultant.
PRACTICAL RESEARCH: Multiple million dollars saved in industry through novel algorithms of my own design, customized for their business needs.
FULL STACK DEVELOPER: Responsible for the entire tech stack at my startup.
EXCELLENT CREDENTIALS: Worked at Palantir, received multiple awards and scholarships.
BROAD AI KNOWLEDGE: I follow recent publications and participate in upskilling programs. I am driven by the desire to understand how thinking works, and this motivated me to get a diverse view of both AI and psychology. I have seen the ups and downs of the industry and can tell the difference between hype and substance.