- Developed an ESG report builder using Retrieval-Augmented Generation (RAG) and open-source LLMs from Hugging Face.- Created a microservice with Flask for seamless integration of the ESG report builder into other applications.- Deployed the solution on AWS EC2 using Docker, optimizing for T4 GPU.- Finetuned an embedding model, improving document classification accuracy by 17.2% for enhanced retrieval results.- Contributed to an open-source RAG project, securing third place in an ITU-organized competition.

Project: Developing a system that helps counting passengers in a buses using computer vision technologies based on deep learning.Workflow:- Detection: we have fine tuned a YOLOv5 and YOLOv8 model on a dataset we have prepared and annotated so the model can detect passengers from the overhead view.- Tracking: using ByteTrack algorithm to keep track on each individual passenger so we can do further analysis.- Counting: We have built and algorithm using OpenCV and python so we can count passengers who gets in and who gets out of the bus.- Data pipeline: Finally, we have built a data pipeline that will get videos from buses and process them into the model that will get us results as a video with predictions and a csv file.