Stefan Todorov

Meet Stefan

1 following

Lead Architect and Systems Engineer building low-latency agentic AI platforms and high-throughput multi-GPU compute systems.

I turn research-grade models into production AI services that are fast, secure, maintainable, and auditable. My work spans fine-tuned LLMs, MCP and gRPC tool pipelines, enterprise vector retrieval, Linux kernel tuning, NUMA placement, real-time scheduling, and low-latency virtualization.

Recent systems work includes a concurrent multi-agent security-analysis platform in C++, Python, and Bash; multi-GPU AMD Instinct and Vega compute infrastructure; and secure Google Cloud agent pipelines using Vertex AI, AlloyDB for PostgreSQL, Cloud Run, and least-privilege IAM.

Core expertise: C/C++, Python, Bash, Linux kernel internals, KVM/QEMU, gRPC and Protocol Buffers, llama.cpp, ROCm/PyTorch, LoRA/QLoRA fine-tuning, Google Cloud, MCP, multi-agent orchestration, and AI security threat modelling.

I also collaborate across engineering teams to operationalise complex systems and mentor engineers delivering reliable AI services.