Backend developer focused on distributed systems and the ML/backend intersection. Working at WEG S.A. — building things on the side that push further.

// portfolio → Luca5Eckert

now

Building VellumHub — solving the read model problem: recommendation serving with zero coupling between the write path and the query path.

Building Kairos — personal knowledge graph engine with dual-pipeline ingestion, local ONNX embeddings on the JVM (tokenization → tensors → inference → pooling), LLM-based OpenIE, and multi-hop retrieval via HippoRAG 2 with pgvector + Neo4j.

Reading Designing Data-Intensive Applications. Studying neural network fundamentals.

interests

Distributed systems and what breaks at scale — consistency models, event-driven architecture, data-intensive design.

The ML/backend intersection — embeddings, retrieval, and inference running on the same stack as the rest of the system.

Neural networks from first principles. How the math becomes the architecture.

projects

VellumHub How do you serve recommendations without coupling the query path to anything? Reactive WebFlux gateway, event-carried state transfer over Kafka, pgvector similarity search with LangChain4j embeddings. Java 21 · Kafka · pgvector · LangChain4j · Redis

Kairos What changes when memory has structure you did not design? Personal knowledge graph engine. Dual-pipeline ingestion with local ONNX embeddings on the JVM and LLM-based OpenIE. Multi-hop retrieval via HippoRAG 2 using pgvector and Neo4j. Java 21 · Neo4j · pgvector · ONNX

OpenIT What does hardware-to-interface actually mean when you build it yourself? IoT access control and payment system — ESP32 over MQTT, SSE for real-time updates, reactive backend. Java 21 · WebFlux · MQTT · React 19

stack

Daily — Java 21, Spring Boot (MVC + WebFlux), Kafka, PostgreSQL, Docker, Redis.

Exploring — ONNX, Neo4j, pgvector, LangChain4j, embedding pipelines, NN fundamentals.

Also — TypeScript, React 19.