Back to projects
AVTOSTOP AI
Backend & AI Engineer
Company
AVTOSTOP Project
Period
09.2024 — 01.2025
Overview
A smart automotive assistant system leveraging advanced RAG pipelines and embeddings for intelligent document retrieval and context-aware responses.
Challenge
Implementing a system that could search through thousands of documents and provide accurate, context-relevant answers using AI.
Solution
Developed a RAG pipeline with embeddings, implemented PGVector for vector storage, built OpenAI model fine-tuning system, and created a Spring Boot API with Keycloak authentication and OAuth2 integration.
Key Highlights
▪Built embeddings pipeline for semantic search
▪Implemented fine-tuning with JSONL datasets
▪Created OAuth2 and Keycloak authentication flows
▪Optimized RAG for accuracy and response time
▪Designed scalable API for high-throughput requests
▪Implemented caching with Redis for performance
Technologies
JavaSpring BootOpenAIPGVectorRAGEmbeddingsKeycloakOAuth2PostgreSQLVector Databases