ALTO AI agent + MCP

01The problem

Device, support and ops workflows at Netsync meant digging through telemetry, docs and internal tools by hand. An LLM could help — but only if it could reach real data without being able to do damage.

02Approach

I built a production AI agent on the Anthropic and OpenAI APIs with LangChain: multi-step reasoning, tool use, memory and guardrails, automating internal device, support and ops workflows.

Then I designed and shipped a Model Context Protocol (MCP) server that exposes structured query and action tools to LLM clients, with secure tool calls into our database, telemetry and internal services.

03Retrieval and evals

Retrieval-augmented generation pipelines with Qdrant and embeddings over telemetry, docs and support content run on AWS.

Fine-tuning runs on S3, SageMaker and Bedrock, with eval harnesses that catch quality regressions before they reach users.

04Results

An agent and tool server in production on top of the same platform that handles 100,000+ concurrent devices — AI that works on live operational data, with guardrails.