ALEXANDRE DNIESTROWSKI

Specialized in STEM domains (Artificial Intelligence, Software Engineering, Computer Science) for advanced LLM evaluation and prompt engineering: 1) Prompt Engineering & Model Evaluation Design advanced prompt strategies to evaluate LLMs across software engineering, computer systems, and distributed architectures Conduct comparative model evaluations (A/B testing) to assess performance differences and identify optimal approaches for specific tasks Develop multilingual (French/English) code evaluation prompts spanning all phases of the software development lifecycle 2) Complex Test Case Development Create expert-level challenges in AI, software engineering, and computer science domains, including PhD-grade reasoning tasks Design edge case scenarios to systematically expose model limitations and failure modes Develop evaluation criteria and scoring frameworks for rigorous quality assessment 3) Quality Assessment & Benchmarking Review and assess model responses against rigorous technical and safety benchmarks Deliver structured evaluations with quantitative metrics and targeted improvement feedback Identify patterns in model failures to inform iterative refinement 4) AI Safety & Systematic Testing Contribute to AI safety frameworks through systematic prompt-based testing and error mode analysis Apply methodical approaches to uncover vulnerabilities and edge cases in model behavior Document findings with precision to support continuous model improvement

Over 25 years of experience in technology development, currently focused on AI-powered application development and advanced language model integration. Leveraging modern AI-assisted development practices to build sophisticated platforms for complex data analysis, intelligent automation, and knowledge management. Key Focus Areas & Implementations: (1) AI-Assisted Development Revolution Transformed development workflow using cutting-edge AI coding assistants (Claude by Anthropic, Gemini/JSC by Google) integrated with Visual Studio Code. This AI-first approach enables accelerated development velocity, intelligent code generation, real-time debugging, and sophisticated architectural guidance - fundamentally revolutionizing productivity and code quality over the past 6 months. (2) LLM Platform Integration & AI-Powered Applications Architected custom integrations with leading AI platforms (OpenAI GPT, Anthropic Claude, Google Gemini) for intelligent data processing and automated workflows. Key implementations: financial investment system combining multiple AI platforms with Neo4j graph databases (Python/Django backend, React frontend); digital humanities research platform leveraging advanced NLP libraries (GATE, NLTK, spaCy, Scikit-learn) for text analysis and statistical processing. (3) Graph Database & Knowledge Management Systems Developed sophisticated graph-based architectures using Neo4j for complex relationship modeling and knowledge graphs. Implementations include text components management system with Python APIs and React interface, optimizing creation and reuse of communication materials through intelligent content management. Technical Stack: Python (Django/Flask), JavaScript/React, Neo4j graph databases, advanced NLP libraries, LLM API integrations (OpenAI, Anthropic, Google), Git/GitHub, CI/CD with GitHub Actions, Google Cloud deployment.

Developed and implemented an AI-powered research and analysis platform that streamlined the processing of complex historical sources through innovative features including a reading accelerator, graph database integration, and statistical analysis tools. Designed and built an intuitive user interface to enhance accessibility and usability of advanced research tools for humanities scholars. Applied expertise in AI prompt engineering to generate precise, actionable outputs across various research applications. Presented research findings at the 16th International Conference on Statistical Analysis of Textual Data (JADT 2022) in Naples, delivering a paper titled "The Augmented Historian".

Directed group responsible for developing and operating intranet/extranet platforms as part of e-Business Metering Division expansion.

Digital transformation of telecom products and software development methods/tools for Philips' European units

Engineering firm specializing in highly reliable software applications. Key contributor with SFENA (now Thales) in the software development of avionic systems: Mirage-2000 and Airbus A300-320 autopilots From single client SFENA expanded customers and industrial sectors: Matra, Thomson now THALÈS; Schlumberger; L’Oréal,...

CRT electronic image processing device connected to a computer system for processing bubble chamber images in nuclear physics. The system is semi-automatic, due to the complexity of the images an operator who can intervene at any time. For the software part, we created a META-MACHINE offering a virtual interface adapted to the physicists conceptual framework where he/she can setup the image processing scripts without any knowledge of the intrinsic computer/software features. The system, dubbed "CRT-Paris" is now on display at the Science Museum La Villette (Paris).