HAYKEL BARGOUGUI

- Developed an agent from scratch in LangGraph, employing OpenAI GPT LLMs and tools for action-based search. This development emphasized advanced prompt engineering techniques. - Utilized LangChain and MultiChain to create scalable model chains that enhance the integration and management of external data sources, such as Tavily search. - Experimented with a variety of LLMs (DeBERTa, LLaMA, BERT, T5, GPT, Phi3) and gained expertise in fine-tuning large models using techniques such as LoRA and QLoRA.

• Designed an AI agentic workflow for industrial scene understanding by integrating spatial information into LLMs. • Processed and aligned 3D point cloud features with instruction-tuned LLMs using LoRA fine-tuning. • Addressed spatial reference resolution, question answering, and scene description to bridge 3D geometry and language. • Trained and evaluated on AWS H100 GPUs, achieving better grounding accuracy and context-aware responses. LLM, 3D-aware LLM, Fine-tuning, Point Clouds, LoRA, AWS Cloud.

- Data collection, data cleaning, data augmentation, and feature selection. - Implemented text vectorization techniques such as TF-IDF, BoW, Word2Vec, Tokenization, Lemmatization, and Stemming. - Developed models including RNN, Random Forest, XGBoost, and Transformer Encoder for sentiment analysis. - Developed and deployed Docker images to Docker Hub, including comprehensive testing and Swagger integration.