Edouard Ducloy

Main contribution and results obtained

Design and deployment of a conversational assistant specialized in Nietzschean philosophy, using RAG (Retrieval-Augmented Generation) technology via the OpenAI API and a vector store system.

Problems solved:

Easier access to a complex philosophical corpus without requiring exhaustive reading of the works

Contextualized answers with precise citations directly from the source texts

Intelligent navigation within dense and often misinterpreted philosophical thought

Objectives achieved:

Creation of a semantic search tool for querying philosophical works in natural language

Successful indexing of a corpus of primary and secondary texts on Nietzsche

Implementation of a relevant retrieval system with similarity scoring

Responsibilities and actions taken

Technical architecture: Configuration of the OpenAI assistant with vector store integration for embedding storage and retrieval

Data processing: Preparation, cleaning, and segmentation of the philosophical corpus into optimized chunks for semantic search

Implementation: Development of query workflows, prompt system configuration, and adjustment of search hyperparameters

Content ingestion: Loading and indexing of Nietzsche's works (primary texts) and secondary studies into the vector store

Testing and validation: Evaluation of answer relevance and adjustment of retrieval strategies

- Design and development of an LLM system with RAG: Responsible for needs analysis, data processing, and design of an artificial intelligence system based on Mistral 7B, using LoRA/QLoRA fine-tuning techniques for a tutoring platform.

- Implementation of a RAG architecture: Setting up a Retrieval-Augmented Generation solution to enrich model responses with specific educational data, thereby improving the relevance and accuracy of interactions.

- Development of advanced educational features: Creation of modules for automatic MCQ generation, student performance analysis, and personalized assistance via chatbot, meeting the specific needs of platform users.

- Model experimentation and optimization: Use of Google Colab for preliminary tests, performance evaluation, and preparation of infrastructure for large-scale deployment.

Technologies Used: Python 3.13, PyTorch, Transformers, PEFT, LangChain, ChromaDB, FastAPI, Sentence-Transformers, BitsAndBytes, MistralAI

- Study, management, and design of an automated testing system: Responsible for needs analysis, project management, and design of an automated testing system for physical products, using proprietary PCBs. This system ensures the quality and reliability of tested products.

- System integration: Supervision of the integration of the automated testing system into the company's internal infrastructure, ensuring optimal compatibility with existing tools and processes. Application of a CI/CD strategy: Implementation of continuous integration and continuous deployment practices via GitLab, enabling fast and reliable delivery of software updates.

- Creation of Python libraries: Development of reusable Python libraries and their deployment on a package server, facilitating their use by other teams and projects.

Technologies Used: Python 3.9, USB ISS, JLink, Flask, Docker, GitLab, Poetry, PyPI