NetBox MCP server and chatbot
Ask your network questions in plain language, answered from NetBox by a self-hosted LLM.
AI & LLM infrastructure
AI and LLM infrastructure for production
Production infrastructure for LLMs and agents: hardened GPU nodes, model serving, retrieval and tool integrations, run with the guardrails and observability that production demands.
What I do
- Kubernetes on hardened Linux GPU nodes: kernel, driver and container-runtime tuning
- GPU-aware scheduling, placement and autoscaling for inference and batch workloads
- Model serving with vLLM and Ollama, with serving controls and deployment guardrails
- Retrieval-augmented generation (RAG) pipelines and vector databases
- Model Context Protocol (MCP) servers that give LLMs safe, read-only access to infrastructure data
- Tenant isolation, self-service APIs, runbooks and observability for LLM platforms
How I approach it
AI workloads are infrastructure workloads with sharper edges: expensive hardware, bursty demand and new failure modes. I treat them with the same rigour as any production system, from capacity planning to on-call runbooks.
Useful AI needs access to real data, safely. I prefer narrow, read-only tool interfaces such as MCP, explicit permissions and full audit trails over broad access.
Track record
Design Kubernetes-based AI infrastructure on hardened Linux GPU nodes with Terraform, Ansible and AWS, including GPU-aware placement and autoscaling.
MercorBuild production LLM and agent platforms using RAG, vector databases, vLLM / Ollama and MCP integrations.
MercorPublished a NetBox MCP server and chatbot that lets a self-hosted LLM answer questions about network infrastructure.
Open source
Related projects
FAQ
Questions people ask
What AI infrastructure does Jad Seifeddine build?
Jad builds production LLM and agent platforms: Kubernetes on hardened Linux GPU nodes, GPU-aware scheduling and autoscaling, model serving with vLLM and Ollama, RAG pipelines with vector databases, and Model Context Protocol (MCP) integrations, with runbooks and guardrails.
What is an MCP server, and has Jad Seifeddine built one?
A Model Context Protocol (MCP) server exposes tools and data to an LLM through a standard interface. Jad has published a NetBox MCP server and chatbot that let a self-hosted, OpenAI-compatible LLM query devices, sites, IP addresses and change history in NetBox in plain language.
Have a hard infrastructure problem?
Open to select consulting, contract and advisory work. Tell me what you are building and where it hurts.