Rosewood AI Training Plan

Overall Read

The plan separates strategic architecture understanding from hands-on technical operating capability.

Kelly should come back able to evaluate the architecture from a systems, information, ontology, source authority, and Rosewood ownership perspective.

Josh should come back able to operate, build, troubleshoot, and extend the technical environment.

The combined goal is not just training. It is to create Rosewood's internal capability to understand, challenge, operate, and eventually extend the agent harness without becoming blindly dependent on outside infrastructure.

Training Logic

Understand the architecture Clarify what exists today, what is roadmap, and where each system fits.

Map source authority Separate current truth, historical information, and generated working output.

Build operating capability Learn how to provision, deploy, inspect logs, manage access, and troubleshoot.

Select first pilot Choose one Rosewood workflow using clean data, low external consequence, and measurable outcomes.

Jody Discussion Points

Area	Question	Why it matters
System boundaries	What exactly are Rosie, Sites, Wayfinder, and Nucleus responsible for today?	Prevents conceptual confusion and bad build assumptions.
Production state	What exists in production today versus roadmap or prototype?	Separates present capability from future direction.
Access	What can Rosewood access, modify, inspect, deploy, and troubleshoot?	Determines whether Josh can become a real first-line operator.
Source of truth	Where do state, logs, prompts, memory, permissions, and model choices live?	Defines traceability and operational control.
Failure modes	What happens when an agent, tool call, queue, or workflow fails halfway through?	Enterprise systems need recovery paths, not just demos.
Portability	Which components are portable if Rosewood later needs to replicate them outside the platform?	Directly affects lock-in and continuity risk.

Gaps To Work Through

Gap	Recommended fix
No explicit deliverable dates	Define interim and final deliverables before training starts.
No named Rosewood pilot workflow	Pick one practical test case: photo exception review, Penmarc install lifecycle, ProVantage kickoff, or scheduling intake.
No access matrix	Ask Jody for a clear access and responsibility matrix.
No source-of-truth map	Require a source-of-truth and freshness diagram.
No economics lens	Add model cost, tool cost, token use, labor savings, quality, and workflow ROI to the outputs.
No security/compliance track	Add data classification, credential handling, audit logging, company boundaries, and approval gate review.

Current recommendation: start with photo documentation and exception flagging as a contained sub-workflow, then attach it to Penmarc or ProVantage depending on which has cleaner data and a more available operating owner.

Strategic Architecture Track

Kelly's Role

Kelly should come back able to evaluate the architecture from a systems, information, ontology, source authority, and Rosewood ownership perspective.

The focus is on judgment: what Rosewood should own, what it can rely on, what creates dependency risk, and how Wayfinder should evolve as an intake layer for the company brain.

Required Outputs

Strategic architecture map.
Wayfinder gap assessment.
Source authority and freshness model.
Rosewood ownership, dependency, and lock-in memo.
Framework for choosing deterministic workflow vs agent vs RAG vs graph vs Site vs human review.

Tier 1 Goals

Goal	Rosewood purpose
Understand Nucleus architecture at a systems level	See how Rosie, Sites, agents, models, tools, memory, data stores, retrieval, graphs, and systems of record fit together.
Understand why the architecture is designed as it is	Learn tradeoffs, dependencies, and reasons, not just components.
Map Wayfinder against company AI-enablement needs	Identify current intake gaps and future information requirements.
Design Wayfinder information collection structure	Improve document, interview, image, workflow, metadata, naming, and source-authority capture.
Understand ontology	Know when entities, relationships, and semantics matter.
Distinguish deterministic, predictive, and generative components	Know where each system mode belongs.
Understand RAG, graph engineering, memory, and systems of record	Choose the right architecture pattern for a given problem.
Understand Sites versus traditional applications	Know when dynamic AI-generated interfaces should replace or supplement traditional software.
Understand current, historical, and authoritative information	Preserve history without confusing old information with current truth.
Challenge technical solutions from the broader system perspective	Ask upstream, downstream, cross-functional, and time-based impact questions.
Assess ownership, dependency, lock-in, and continuity risk	Identify what Rosewood should own versus rely on outside infrastructure for.

Tier 2 Goals

Compare the architecture with other frontier agent architectures.
Develop a framework for software vs deterministic workflows vs single agents vs multi-agent systems vs RAG vs graphs vs Sites vs human review.
Think through AI-native program management with many agents and builders.
Evaluate Wayfinder as the intake layer for a broader company brain.
Identify how historical decisions and workflows become reusable intelligence.

Technical Operator Track

Josh's Role

Josh should come back able to operate, build, troubleshoot, and extend the technical environment.

The focus is on practical control: environment access, deployments, logs, state, permissions, model routing, tool-using agents, graph structures, Cruxos reporting access, and portability.

Required Outputs

Technical environment runbook.
Working example of a tool-using agent.
Working example of a multi-agent or evaluator pattern.
Practical graph or relationship-based retrieval example.
Wayfinder ingestion prototype or design.
Secure read-only connection plan for Cruxos reporting database access by Rosie.
Portability assessment.

Tier 1 Goals

Goal	Rosewood purpose
Become technically functional in Rosie	Operate, configure, and troubleshoot common issues.
Understand the Nucleus technology stack	Know how agents run, models are used, tools connect, and environments are structured.
Become functional in the development environment	Provision environments, deploy work, inspect logs, manage access, and locate code, data, and state.
Build multiple agent patterns	Single agent, tool-using agent, coordinator, specialist, evaluator, and multi-agent workflows.
Build with multiple models	Understand model routing by cost, complexity, speed, and quality.
Learn multi-agent and colony architecture	Understand coordination, handoffs, inspection, and concurrent operation.
Learn practical graph engineering	Build graph structures where relationships matter.
Build Wayfinder technical intake	Ingest documents, interviews, images, structured data, and other source material.
Handle metadata, naming, versioning, historical storage, and retrieval	Keep information organized, current, traceable, and recoverable.
Connect Cruxos reporting database access to Rosie	Establish a secure, read-only path for Rosie to query ERP/reporting data with approval gates, logging, and source traceability.
Build agents that gather missing information	Move beyond passive intake to gap identification and follow-up.
Build computer/browser operating agents	Automate work where clean APIs do not exist.
Solve a non-linear workflow problem	Handle ambiguity, multiple systems, state, exceptions, and judgment.
Understand replicable vs Nucleus-dependent capability	Distinguish platform dependency from portable capability.
Become Rosewood's first-line technical operator	Diagnose, fix routine issues, and know when Jody or Nucleus must be involved.

Tier 2 Goals

Move or help move Rosie onto the Nucleus environment if feasible.
Build one real Rosewood workflow using computer-operating agents.
Build a working multi-agent process with evaluation and human escalation.
Build a dynamic Site-style interface against real data.
Experiment with model routing based on cost, complexity, speed, and quality.
Build a contained Cruxos-related prototype only after higher-priority training goals are progressing.