Portfolio skill upgrade matrix · August 2026
The canonical project is not a certificate and not a pile of demos. It is the visible bridge from twenty years of production recovery to agentic AI delivery: diagnose the floor, stand up the path, measure it, and leave the next operator a system they can run.
I was not practicing incident response. I was parachuted into live Fortune-100 data platforms where the clock and the customer were both present.
The old loop survives the stack change: establish a baseline, name the bottleneck, eliminate it, repeat. A louder tool is not a diagnosis.
The lab makes the agent path inspectable. MCP sits on :8001; UIs share one OpenAI-compatible contract; the engine stays running.
One compose. One network. One GPU. qwen is the engine. The architecture rule is itself part of the demonstration.
CDC, BDM, Grid/HA, shared filesystems, scheduler behavior and database waits are not vocabulary words. They are where the symptom lies.
The same systems thinking appears in old and new clothes: Grid/HA for enterprise integration; UNIFIED for private inference.
A four-agent runtime isolates processes and hands work across Pydantic-enforced JSON. Validation is a loop, not a hope.
Local Grid Suite turns the GPU claim into telemetry: discovery, benches, routing, SQLite logs and repeatable reports.
Leadership is not a title. It is the runbook, the QA twin and the person who can operate after the closer leaves.
FDE work is familiar terrain: enter someone else’s stack, hear what is actually broken, explain the path without notes, and ship something usable.
n8n ingests live metrics, retrieves prior context with RAG, writes Discord updates, and only fires WhatsApp when a threshold or Rank 1 gate is met.
The work is packaged as live GitHub + Hugging Face twins: a recruiter can click the stack, the evidence and the application-specific packet.
Twenty years of delivery is the credential: production competence, warehouse/domain depth, systems thinking, communication, leadership and customer fit. Do not dilute those with “I write clean code.” Name the floor and the receipt.
Agentic AI is the lab year. Not a certificate: a stood-up path from prompt to tool to model and back, with MCP, contracts, telemetry, gates and packets someone else can inspect.
Day 1 chat path. Day 2 coding pack. Day 3 memory and workflow. The next 18 days are repetition until you can teach the stack without notes, with qwen still up.
From Informatica recovery → private RTX lab → UNIFIED → MCP → packets → hunt board. The portfolio is a trail of operational evidence, not a mood board.