Who I Am
I am Karim Rahimov. Passionate software engineer specializing in creating innovative solutions. Looking to join a dynamic team.
Featured Projects
Code Showcase
Project: Custom Reasoning SFT Layer Platform for Open-Source Models
Designed and developed a platform enabling users to train a custom Supervised Fine-Tuning (SFT) layer on top of open-source models (e.g., LLaMA) to transform them into O1-type models with enhanced reasoning capabilities.
Leveraged Reinforcement Learning (RL) and Low-Rank Adaptation (LoRA) techniques to efficiently train and integrate the additional reasoning layer into existing open-source models.
Optimized model performance by fine-tuning hyperparameters and ensuring seamless integration of the SFT layer with the base models.
Demonstrated the platform's effectiveness by improving model reasoning and decision-making capabilities, making it suitable for complex problem-solving tasks.
Latest Work
Data Scientist
General Motors, Chicago
Engineered robust data pipelines using Apache Spark to process and clean 500GB+ of vehicle sensor data daily, reducing data preprocessing time by 60%
Developed and optimized CNN models for autonomous vehicle perception, achieving 95% accuracy in real-time object detection and classification
Built and deployed real-time data streaming pipeline integrating multiple sensor inputs (LIDAR, camera, radar) for autonomous driving system, processing 100K+ data points per second
Designed and deployed data quality monitoring systems using Apache Airflow, reducing model retraining frequency by 40% through improved data validation
Software Engineer
Stonewolf Contracting, New York
Implemented user interface with NEXT.js, resulting in improved order efficiency and increased customer base by 12%.
Developed a fast and efficient order flow system using Node.js and REST API within the backend, which automated the order flow of the company’s products, resulting in an 11% increase in revenue.
Created a database schema using PostgreSQL, which stores the data about every customer and the product that they ordered.
Project: Scalable MLops Platform for Data Analytics and Model Management
Built a distributed, cloud-native MLOps platform inspired by BigQuery, enabling petabyte-scale data analytics and ML workflows.
Designed a columnar storage engine using Apache Parquet and an SQL-like query engine with Apache Calcite for optimized query execution.
Integrated Apache Spark and Dask for distributed computing and Kafka/Flink for real-time data ingestion and streaming analytics.
Implemented a model registry with MLflow and automated retraining pipelines using Kubeflow, deployed on Kubernetes for scalability and fault tolerance.
Reduced query execution times by 40% while ensuring secure data access via RBAC and compliance with governance policies.