Narges Norouzi
Data Scientist/Computer Vision Engineer
About Me
Data Scientist/Computer Vision Engineer with 5+ years of experience in researching and building applied machine learning, computer vision, and big data applications distributed focused web crawlers, and overcoming high load real-time stream processing issues in different businesses. Experienced at designing, developing and deploying machine learning and deep learning models, as well as proficient in predictive modeling, data processing, and data mining algorithms. Finally, I am passionate about cutting-edge technology, researching and solving real-world problems with previous experience in this field.
Research & Work Experience
* Project1: SnapMode ((Iranian Large-Scale Image Retrieval Engine) See
- Designed and developed a Low Latency Scalable Focused Web Crawler to extract fashion data from E commerce websites using Apache Storm, Solr, Kafka and Milvus. (+10M Product Pages).
- Enhanced the content-based image retrieval Accuracy using a Triplet Generative Adversarial Networks (CBIR-GAN) to feature embedding.(82% accuracy on the in-shop products).
- Optimized the search performance of vector queries using clustered milvus.
* Project2: Water Sensor Data Analysis See
- Designed and developed a real-time big data architecture for water sensor data ingestion, fusion and analysis using transformer-based deep learning algorithms
*Project3: Enamad (E-commerce Websites Monitoring System) See
- Built a fake logo detection system that leveraged transfer learning using pre- trained Resnet-18 model to successfully identify real Enamad logo in Iranian E-commerce websites.(92% accuracy)
*project4: BigMehr(AI-Enabled Marketing Platform)
- Designed and Developed an AI-Enabled marketing system based on Horton- works big data platform to process and analyze the customer data of a USSD application (#780) with +3M active users.
- Developed a customer churn prediction model using LSTM networks that im- proved monthly retention by offering discount and running relevant marketing Campaign.
- Improved marketing team process, resulting in a 30% decrease in time needed to infer insights from customer data used to develop marketing strategies.
- Used Prophet algorithm to forecast daily company sales with a 85% accuracy rate.
*project5: Sepahtan(Driver Behavior Monitoring System)
- Developed a driver behavior monitoring system using Hortonworks big data platform with +3k driver.
- Implemented driver behavior clustering model using Apache Spark MLlib kmeans algorithm.(Silhouette with squared euclidean distance = 0.75)
* Project6:Rasad (University News Analysis and Tracking System)
Designed and developed an news analysis and monitoring system that leveraged from BERT model for sentiment analysis and improved negative comments detection with 82% accuracy rate.(Focusing on university news)Portfolio
Content-based Image Retrieval /Apache Storm / Solr / Kafka / Milvus / Color Detection/ CNN Models / FastAPI/ Vuejs
SnapMode:Iranian Large-Scale Image Retrieval Engine
Designed and developed a Low Latency Scalable Focused Web Crawler to extract fashion data from E-commerce websites. Enhanced the content-based image retrieval Accuracy using a Triplet Generative Adversarial Networks (CBIR-GAN) to feature embedding. Product Link, Paper Link
Deep Generative Adversarial Networks/ Self-supervised Image Representation/ Contrastive Learning/ Triplets
Cbir-Gan: A Triplet Generative Adversarial Network for Content-Based Image Retrieval
This project proposes a triplet generative adversarial network (GAN) based on the idea of integrating deep metric learning methods with GANs to benefit from the advantages of both at the same time. In this model, three generator networks that use a triplet loss function are responsible for learning a similarity measure over objects and embedding images in an appropriate vector space. In these networks, a CNN-based perceptual loss function is also employed to force the generators to adhere a certain type of structural features in intermediate layers. Since only triplets must be used as network inputs in the proposed method, the learning process is performed in a semi-supervised way. Product Link, Paper Link
Apache Storm / Solr / Sentiment Analysis / Twitter & Web Crawling / Bert NLP Model / FastAPI / Vuejs
Rasad (University News Analysis and Tracking System)
Designed and developed a news analysis and a monitoring system that leveraged from the BERT model for sentiment analysis and improved negative comments detection with 82% accuracy rate.(Focusing on university news)
GANs models / Wavelet Transform / Pytorch
Texture synthesis in image to image translation in the field of fashion AI
In this research, we presented a generative model called WBT-GAN for texture synthesis problem, which was an extension of the existing Texture-GAN network using a four-level wavelet transform and error definition based on it in the objective function of the model.
Education
Thesis: Detection of changes in Longitudinal MRI images
GPA : 19.04 (A+)Thesis: Analysis, Design and Implementation of Student Information Management System
GPA: 17.03
Publications
2021.
2021.
in Journal of Medical Signals and Sensors (JMSS), pp. 156-164, Vol. 1(3), (2011).
Journal of Medical Signals and Sensors (JMSS), vol. 1(2), pp. 138-148, (2011)
JJournal of Medical Signals and Sensors (JMSS), Vol. 3(2), pp. 92-103, (2013)
