GenAI Engineer building production-grade multimodal RAG systems — end to end, from ingestion to deployment.
I specialize in finance-domain retrieval-augmented generation: systems that have to get numbers right, resist prompt injection, and prove their accuracy with a real evaluation harness instead of a demo.
MAGIK — a multimodal, agentic RAG assistant that ingests text, PDF, DOCX, XLSX, images, audio, and video, routes queries through an agentic controller, retrieves with a hybrid BM25 + dense pipeline, and verifies its own answers before they reach the user. Runs entirely on open-source models — no paid LLM API in the request path.
See it live → vk-ai.online