Hi, welcome to my website! :)
Research Interest
I work on graph neural networks over relational databases. More than 70% of operational data in finance, healthcare, and enterprise systems lives inside relational schemas, yet extracting predictive insight from it still requires manual SQL and feature engineering. I am working toward systems that learn directly from relational structure — treating tables, foreign keys, and entity relationships as a heterogeneous graph — and expose that signal through natural-language interfaces. Text-to-SQL tools like Databricks Genie are meaningful progress, but they translate queries into lookups rather than learning from the relational structure itself, leaving the predictive signal embedded across tables untouched. Closing that gap is the frontier I am building toward.
Experience
I'm a Backend & AI Engineer at Dcentralab Ltd, an Israel-based Web3 infrastructure company, where I build production multi-agent LLM systems. On Traia I redesigned the Modifier and Drafter services from barely functional to production-ready, built the template-generation and agent-to-agent communication systems that enable autonomous multi-agent trading, designed the execution engine that runs on live capital, and published the traia-iatp Inter-Agent Transfer Protocol across 111 versioned releases. On Web3Index I was primary AI engineer for a production AI search and chat system over 64,000 Web3 and crypto projects — semantic search, multi-turn conversation, and persistent chat history for live users.
Independently, I designed and built the Branham Model API, a production hybrid agentic retrieval system combining BM25 and dense vector search over 207,000 paragraph-aware chunks from 1,203 sermons, with conditional neural reranking, Reciprocal Rank Fusion, and SSE streaming. Deployed at branhamsermons.ai with zero marketing, it has served 12,960 requests across 59 countries in 30 days.
On the research side, I implemented GCN, GraphSAGE, GAT, and knowledge graph embedding architectures under the informal guidance of Prof. Pietro Liò (University of Cambridge, Computer Science), and completed Stanford's graduate-level Machine Learning with Graphs course. I am co-first author on peer-reviewed work in Energy Reports (Elsevier, 2024) and Localized Energy Transition in the 4th Industrial Revolution (Taylor & Francis, 2024), and co-first author on a Vision Transformers in Medical Imaging review on arXiv (currently ~100 citations). Five SCIE/Scopus-indexed journals have independently invited me as a peer reviewer.