David HUANG

- Led the design and deployment of a production-grade legal AI assistant leveraging RAG pipelines and agentic architectures (OpenAI, Gemini, Mistral), improving access to legal information at scale - Built multi-stage retrieval systems combining key-word and vector search with Elasticsearch, with re-ranking pipelines to optimize relevance and latency - Designed and optimized RAG architectures (chunking strategies, embeddings, retrieval orchestration, source attribution) for high-precision legal use cases - Implemented evaluation and benchmarking frameworks for LLM systems (retrieval quality, answer accuracy, latency), driving iterative product improvements - Led A/B testing and experimentation on systems relevance - Deployed and scaled models in production with robust MLOps pipelines (MLflow, Vertex AI, SageMaker, DVC), ensuring monitoring, reliability and observability - Trained and fine-tuned NLP models (BERT, HuggingFace) for legal tasks, including knowledge graph construction and domain adaptation - Contributed to technical strategy and AI watch, including LLM optimization techniques (fine-tuning, distillation, RLHF) and emerging frameworks (vLLM, DSPy, langgraph, literalAI)

Co-organize Paris’s NLP meetup: selecting trending topics, sourcing speakers, and running events.

Project — Data engineer @ TotalEnergies | Energy Forecasting Tool Context Developed a machine learning-based tool to forecast energy consumption and optimize operational costs across industrial processes. Challenges Handling large-scale, heterogeneous time series data Data streaming ingestion and processing pipelines Technologies Used Time series analysis, Azure function, Azure Cloud Event Hub, Spark Databricks, Flask, PostgreSQL, Pandas, Scikit-Learn Project — Data engineer @ TotalEnergies | Hydrogen Optimization Platform Context Built a real-time decision-support system for hydrogen production and consumption optimization, enabling better resource management and operational efficiency. Challenges Designing real-time data streaming architecture Providing accurate forecasting for production and consumption Delivering actionable insights through intuitive dashboards Technologies Used Azure Cloud Event Hub, Azure function, Flask, PostgreSQL Project — Data scientist @ Faurecia | Visual Inspection with Deep Learning Context Developed a computer vision solution for automated visual inspection in an industrial production environment, aiming to improve quality control and reduce manual intervention. Challenges Detecting subtle defects in real-time under industrial constraints Ensuring robustness across varying lighting and production conditions Integrating the solution into production systems Technologies Used Keras, OpenCV, signal processing techniques, production deployment pipelines Project — Data scientist @ Airbus | Defect Detection Architecture Context Designed a scalable and optimized architecture for automated defect detection in an industrial setting, leveraging deep learning for high-precision analysis. Challenges Building a scalable inference architecture for large volumes of data Achieving high detection accuracy on complex visual patterns Technologies Used YOLOv3, TensorFlow

Consulting Assignments | Luxury & Cosmetics Industry Conducted statistical analysis and customer data valorization using SQL Identified actionable business levers from purchase behavior data Delivered ROI-focused analysis of marketing campaigns Performed statistical feasibility studies for loyalty and rewards programs

Image Dataset Creation - Web scraping using Python and BeautifulSoup Data Labeling - Automatic classification of unlabeled images using Keras and TensorFlow Edge Deployment (Jetson TX2) - Configuration and installation of deep learning dependencies (CUDA, cuDNN, Caffe)

National entrepreneurship program, “Les Entrepreneuriales” - Launched an e-commerce purchasing platform for student organizations.