Kaspars Polis

AI-Powered Product Discovery & Visual Search Platform

Designed and led development of an AI-powered product discovery platform combining semantic search, image embeddings, vector databases, computer vision, and retrieval-augmented generation.

The platform supports multimodal search across product catalogs using text, images, and metadata while providing dynamic faceting, AI-powered ranking, and automated data enrichment.

Built end-to-end architecture including data pipelines, vector search infrastructure, backend services, AI models, and frontend integration.

• · Spearheaded industrial-scale computer vision projects and production ML pipelines that processed thousands of daily orders and enhanced operational efciency across fulfllment systems.

• · Automated quality-control pipeline with computer vision, replacing an eight-person manual QC workfow and achieving 90% inspection automation

• · Established end-to-end ML pipelines for training, evaluation, and production deployment on AWS SageMaker

• · Developed multimodal embedding systems for semantic search, RAG workfows, and vector search using pgvector, FAISS, and OpenSearch ANN

• · Optimized AI inference pipelines for latency reduction and high throughput

• · Led architecture decisions for ML systems and mentored engineers on model development and deployment practices

• · Researched and implemented prototypes for AI-assisted image generation and prompt optimization, integrating mechanized QC enhancements into pilot workfows.

• · Built reinforcement learning prototypes for intelligent trafc control using SUMO simulations

• · Integrated ML models into urban mobility platforms, improving prediction accuracy by 25%

• · Developed backend APIs, automated testing frameworks, and data pipelines