Carlos AlvaradoTechnology Solutions Architect
Contact

Conversational network operations

Ask infrastructure questions in natural language without granting the model direct device access.

My role
MCP tool & interface builder
Context
Network operations · Local AI
Project status
Client beta

Beta scope

  • Betasmall client group
  • Localinference boundary

A small client beta. Every configuration change requires human approval; no measured escalation reduction is claimed.

Conceptual illustration · not a client topology

Inside the work

How I approached it

  1. Ask

    An engineer asks a network question through a local LLM in Open WebUI.

  2. Mediate

    Custom, scoped MCP tools sit between the model and live network equipment.

  3. Interpret

    The shared normalization layer makes device outputs consistent and interpretable.

  4. Approve

    Every write operation requires explicit human confirmation before reaching a device.

Decisions & trade-offs

Architectural reasoning based on the supplied project account.

No direct device reachTool-mediated interaction

Tool-mediated interaction

The model can only interact through libraries built and scoped for this purpose.

The trade-off

New capabilities require explicit tool design rather than unrestricted command generation.

Separate reads from writesHuman confirmation for every change

Human confirmation for every change

Read operations support investigation. Configuration changes must be confirmed before execution.

The trade-off

The approval step is deliberate friction around operational consequences.

Validate before wider useSmall client beta & internal environment

Small client beta & internal environment

Use cases are validated internally before being exposed to production networks.

The trade-off

Beta scope should not be presented as broad production maturity.

Technologies by purpose

Conversation & inference

  • Open WebUI
  • Ollama
  • Local LLM

Controlled network operations

  • Python
  • MCP
  • Cisco
  • Scoped tools
  • Human approval
  • Normalized device output

Scope & evidence

About this case

Project description supplied from LinkedIn. Client infrastructure and operational commands are not exposed here.

Carlos on LinkedIn

Project account

The interface broadens access to network investigation and is being validated with a small number of clients. No measured escalation reduction or autonomous remediation claim is made.

Owner-reported account. Private client configurations and original architecture documents are not published with this case.