Rani Baghezza

Experience

• 

  Nexter Systems

  Jun 2013 - Jul 2013

  Assembly line worker
• 

  IHMTEK

  Jul 2015 - Sept 2015

  Video game developper

  Specifications and development of a mobile serious game (Ludomuse), using C++, Java and Cocos2dx. Cross-plateform and cross-language project (Java/C++ using JNI).

• 

  IHMTEK

  May 2016 - Aug 2016

  Software Developper

  Specifications and development of a mobile app with Unity (for Android/iOS) in C#.Specifications (UML, interface, improvements...), front-end/back-end interface (parsing, requests...), MVC architecture.

• 

  CMRE Logiciel

  Jan 2017 - Sept 2017

  Data Scientist

  Big Data pipeline construction (Hadoop, Spark)Exploration of semi-structured databases (MongoDB)Implementation of a Random Forest model for regression and predictions

• 

  UQAC

  Aug 2018 - Aug 2022

  Doctoral Student

  Pedestrian Profile Recognition using Sequences of Low Resolution Thermal Images and CNN-BGRUDeep expertise in low-level Tensorflow, and hyperparameter optimizationDesigning and building prototypes to capture thermal images on Arduino and Raspberry PiLiterature review and scientific publication in high-impact journals (IEEE Internet of Things)

• 

  Université de Sherbrooke

  Aug 2022 - Jan 2024

  Postdoctoral Researcher

  Real-Time 3D Object Recognition and Tracking in Mixed RealityVision model fine-tuning (SAM, Detic, FastSAM, YOLO).Python development (multithread, websockets, vectorized functions, 3D clustering)PhD students mentoringScientific publishing

• 

  Rani AI Solutions & Engineering

  Aug 2023 - now

  AI Engineer

  -Built an AI Assistant (LLM+RAG) in Azure on 100Go+ of data to speed up the work of market researchers for VELOX Operations [Azure, Cloud, RAG, ETL, Python, JS, React]=> R&D work around AI assistants in a technical field. Improved information gathering by an estimated factor of 3x-Built an LLM+RAG pipeline to convert PDF invoices into structured JSON invoices for Nuxly & Antarctic Foods Aquitaine. [Python, LlamaIndex, Pinecone, RAG]=> Significant automation of the invoice processing pipeline.-Built an MLOps pipeline with embedded deployment of a vision model (EfficientDet-Lite) and OCR on RaspberryPi for Genalyte [Python, Tensorflow]. => Improved error detection and parsing of the error messages in a closed medtech environment.-Optimized, quantized, and ran embedded vision models for Survue AI (MobilenetV2-SSD and YOLOX on EdgeTPU and TI AM62A7 chip) [Python, PyTorch, ONNX]=> Multiplied vehicle detection accuracy by 2x and explored complex quantization as well as accuracy/speed optimization for a real-work embedded product. Mentored a junior student on ML practices.-Fine-tuned an object detection model (YOLOv8) and built an ETL pipeline to detect stock patterns on Yahoo Finance and Coinbase stocks [Python, API, ETL, Data Annotation]=> 80% + accuracy in simple pattern detections on stock charts (Double top, double bottom, heads and shoulders...)-Trained LLM models using RLHF for DataAnnotation-Built a prototype of a Python RPG game that uses a LLM+RAG approach to generate quests in real-time-Consulting for clients on vision projects (Depth estimation, computer vision)-Chess AI project using CNN to train an AI to play against itself or a human.-Mentoring for ML students Show less