Constantin Stipnieks

As the first hire, I played a key role in designing and building all core systems from the ground up ●​ Developed an LLM-based chatbot to provide users with insights into a comprehensive catalog of product attributes and generate personalized recommendations ●​ Fine-tuned embedding models and utilized vector databases to build a scalable recommendation engine that retrieves similar images with user-customizable criteria ●​ Built and deployed an image attribution pipeline using Vision-Language models, continuously improved through prompt engineering and few-shot learning ●​ Implemented a modular data pipeline aggregating and processing scraped data from multiple sources ●​ Created efficient database schemas and optimized complex SQL queries to enhance speed and scalability.

●​ Conceptualized and developed a training and inference pipeline performing object detection and semantic segmentation with high accuracy on very large images containing thousands of objects ●​ Reduced post-processing time by 99% through parallelization and algorithmic optimization ●​ Researched and implemented a model to segment changed areas in multi-temporal image pairs, improving F1-score on benchmark dataset from 80% to 88%.

●​ Researched and developed methods for high-resolution 3D reconstruction of a scene from depth images ●​ Increased resolution of the 3D reconstruction application by 400% through algorithmic improvements.