Carlos AlvaradoTechnology Solutions Architect
Contact

AI-assisted infrastructure assessments

Structured findings and risk scoring from standardized data collection, with engineer review.

My role
Platform designer & builder
Context
Enterprise infrastructure · Local AI
Project status
In use · Expanding

Reported effort & adoption

  • ~45 → ~20 hreported assessment effort
  • 5engineers using it
  • 4client assessments completed

Approximate owner-reported effort, not a controlled benchmark. Adoption figures reflect the supplied project account.

Conceptual illustration · not a client topology

Inside the work

How I approached it

  1. Collect

    Read-only MCP tools use Netmiko and NETCONF with a defined command set.

  2. Normalize

    Convert device output into one consistent schema before it reaches the model.

  3. Analyze & interview

    Combine vulnerability analysis with a structured interview that reveals operational gaps.

  4. Review & report

    Engineers review findings, risk/impact priorities and remediation recommendations.

Decisions & trade-offs

Architectural reasoning based on the supplied project account.

Keep inference localOn-premise LLM

On-premise LLM

Local processing was a requirement for client environments belonging to financial institutions.

The trade-off

Data boundaries take priority over the convenience of an external inference service.

Control the model inputsDefined collection & normalization

Defined collection & normalization

Scoped tools and structured outputs reduce inconsistent context reaching the analysis layer.

The trade-off

The collection and schema need deliberate engineering; the model cannot substitute for that foundation.

Keep the engineer accountableEngineer-in-the-loop review

Engineer-in-the-loop review

The tool structures analysis. The engineer owns the judgment and validates the findings.

The trade-off

Human review remains necessary, particularly for hallucination control and remediation decisions.

Technologies by purpose

Local analysis & tools

  • Python
  • Ollama
  • MCP
  • Local LLM
  • Risk matrix

Collection & normalization

  • Netmiko
  • NETCONF
  • Read-only credentials
  • Defined command set

Scope & evidence

About this case

Owner-reported figures from the supplied LinkedIn project description. Approximate effort comparison, not a controlled benchmark.

Carlos on LinkedIn

Project account

The supplied account reports a fifth assessment in progress and findings that manual review had missed. Remediation documentation and scoped proposals are the next extension, not a completed capability.

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