Amir Masoud Sefidian Email & Phone Number

Berlin, De

Berlin, Germany

- Designed and implemented a deep learning-powered audio source separation web application that enables users to separate a song into high-quality audio stems in less than one minute instead of manual and imperfect conventional methods.- Established Sefidian Academy, with 12K+ monthly users, dedicated to explaining concepts in machine learning and related topics.- Created a service utilizing PySpark and NLP techniques to detect fake/real job postings on Indeed; attained an F1 score of 0.9.

(Formerly ParticleB.ai)Worked on different components of an AI-based Algorithmic Trading System:– Reduced the runtime of the walk-forward optimization process by 83% by developing "A walk-forward optimization framework for algorithmic trading strategies on cloud architecture".– Implemented a parameter optimization service that periodically finds optimal parameters for the algorithmic trading engine. Increased Rate of Return (RoR) of the portfolio up to 2%.– Contributed to developing a Backtesting Framework that provides various tools for implementing and testing trading strategies such as fetching data, extracting features, training predictive models to generate signals, allocating assets, sizing the bets, executing orders, and evaluating portfolio performance.Designed and developed FIDIBO (https://fidibo.com/) Recommendation System (the largest digital platform for Ebooks, Audiobooks, and Podcasts in Iran):– Helping FIDIBO users discover new and personalized items by developing content-based, collaborative filtering, sequence-aware, and hybrid recommendation models. – Enhanced offline performance metrics NDCG@5, MAP@5, and Coverage by 50%, 33%, and 150%, respectively.– Performed online evaluations:a) A/B tests to measure the effect of different recommendations on business metrics. Increased CTR by 22% compared to the baseline that recommends popular items.b) Funnel analysis to trace how users interact with real-time recommendations during their journey throughout the application. Boosted the Conversion Rate for ”Complete purchase” and ”Add item to favorites” actions by 14% and 8%, respectively.– Utilized Airflow to build data preparation, embedding generation, batch recommendation prediction, and post-processing pipelines.– Created interactive recommendation evaluation and exploration dashboards using Plotly Dash.

- Developed a system that gives insights into the metrics of an organization using ML algorithms. It integrates metrics of an organization as time series and then performs analytical tasks such as prediction, segmentation, trend analysis, outlier detection, metric importance analysis, and descriptive analysis. Tools: Python, Apache Kafka, Redis, Docker.• Created reports and visualizations to summarize and communicate the findings of data analysis to stakeholders.• Implemented customized machine learning algorithms to predict future trends and outcomes based on historical data such as product demand.• Tracked key performance indicators (KPIs), including customer satisfaction, product sales, and packaging costs, over time to identify trends and areas for improvement.

- Built several AI-based services using PyTorch: – Automatic Image Captioning service by training an Encoder(CNN)-Decoder(LSTM) network on the MS COCO dataset. – Facial Keypoint Detection service that recognizes locations of 68 facial keypoints in an image using deep CNNs. – Sentiment Analysis service for Amazon customer reviews dataset using deep RNNs (91.7% accuracy on the test dataset). – Energy Consumption Prediction service using GRU/LSTM networks.• Designed and launched a web application to efficiently manage 40K+ appointments, 9K+ transactions, and 4K+ users at an institution.• Proposed and developed three novel missing value imputation approaches in the field of data preprocessing using machine learning.

Full-Stack Web Developer, Software Quality Assurance, and Control Tester

C/C++ Developer for AVR Microcontrollers

Master'S Degree, Computer Software Engineering, Gpa: 4.0/4.0

Thesis: "Improving missing value estimation and inconsistencies detection using data partitioning techniques" First rank in the class.

Bachelor Of Science (B.Sc.), Computer Software Engineering, Gpa: 3.96/4.0

Thesis: Designing an online consultation system. First rank in the class.

High School Diploma, Mathematics & Physics, Gpa: 19.75/20

First rank in the class