At Levit8 Technologies, I built Levit8 Recruit’s agentic outreach system, which places calls, sends emails, and texts candidates using LLM-driven tool use.
I also integrated speech-to-text, text-to-speech, and voice activity detection for AI-led video interviews, cutting first-round screening time by 80%. I built semantic search over CVs and recruiter materials using MongoDB Atlas Vector Search.
For Levit8 CX, I built a Qdrant-backed RAG pipeline grounded in customers’ SOPs, FAQs, and policies. Its AI-and-human handoff auto-resolved 65% of routine tickets.
At ICECREAM LABS, I worked on image-generation and editing pipelines, including Stable Diffusion fine-tuning, and fine-tuned an instruction-following LLM. Earlier, at the University of Groningen, I built a neural-network classifier for Parkinson’s and Alzheimer’s using FDG-PET images.
