How EA Studio builds an Enterprise Architect + CTO master brain

This is the execution trace of your CrewAI-powered decision system: how intent is routed, how prior decisions are recalled, how multi-agent debate is governed, and how every run becomes reusable enterprise intelligence.

Total runs

Fast Draft / Committee Debate

2 / 0

HITL flagged

Learning entries

A 0 · R 0 · P 1

TOGAF / ArchiMate View

Simplified layered view for readability: business outcome, orchestration services, and evidence services.

%%{init: {"flowchart": {"htmlLabels": false}} }%%
      flowchart TB
      subgraph L1[Business Layer]
        A1[Stakeholder Ask]
        A2[Governed Recommendation]
        A3[Approved Architecture Decision]
      end

      subgraph L2[Application Orchestration Layer]
        B1[Intake Service]
        B2[Routing Service]
        B3[Master Brain Service]
        B4[Lane A Draft Service]
        B5[Lane B Committee Service]
        B6[Validation and HITL Service]
      end

      subgraph L3[Technology and Evidence Layer]
        C1[Runtime Platform\nPython and FastAPI]
        C2[Context Services\nRAG and Policy]
        C3[Provenance Hash Chain]
        C4[Learning Memory]
        C5[Artifact Pack\nADR, WAF, Ref-Arch, Framework, Options, Roadmap, Migration, Integration, Landing-Zone, RAG-Pattern]
      end

      A1 --> B1 --> B2 --> B3
      B2 --> B4
      B2 --> B5
      B4 --> B6
      B5 --> B6
      B6 --> A2 --> A3
      A3 --> C5

      B1 --> C1
      B2 --> C1
      B3 --> C2
      B4 --> C1
      B5 --> C1
      B6 --> C3
      C5 --> C3
      C3 --> C4
      C4 -. informs next run .-> B3

Download this schematic directly as SVG or PNG.

Five Committee Crews · Clear Flow

Each Lane B committee crew is multi-agent, not single-agent: advocates + guardrails reviewer + judge.

Step 1: Framework Selection Step 2: Migration Target Step 3: Integration Pattern Step 4: Landing Zone Step 5: RAG Pattern

Crew	Decision Domain	Agent Composition	Primary Output	Code Path
Framework Selector	Which orchestration framework to standardize on.	6 advocates + 1 guardrails + 1 judge = 8 agents	Framework recommendation and weighted scorecard (ADR-ready).	lane_b/framework_selector.py
Migration Advisor	Best Azure runtime target for workload migration.	6 advocates + 1 guardrails + 1 judge = 8 agents	Migration target recommendation and trade-off matrix.	lane_b/migration_advisor.py
Integration Pattern Selector	Best integration style across API, events, queue, workflow, stream, federation.	6 advocates + 1 guardrails + 1 judge = 8 agents	Integration pattern recommendation and risk findings.	lane_b/integration_pattern_selector.py
Landing Zone Advisor	Best network and governance topology for deployment.	6 advocates + 1 guardrails + 1 judge = 8 agents	Landing-zone topology recommendation and governance view.	lane_b/landing_zone_advisor.py
RAG Pattern Selector	Best retrieval pattern for quality, cost, latency, and guardrail fit.	7 advocates + 1 guardrails + 1 judge = 9 agents	RAG pattern recommendation with scorecard and guardrails findings.	lane_b/rag_pattern_selector.py

Meet the Agents at Work

Intake Mapper

Mission Framer

I translate a noisy ask into a crisp mission so every downstream step starts aligned.

web.app::generate

Route Strategist

Path Selector

I pick the right decision lane quickly so teams spend effort where risk and impact are highest.

router::route

Master Brain

Memory Curator

I bring proven precedent into the room so your team does not keep paying to relearn old lessons.

brain::build_brain_context

Debate Crew

Multi-agent panel

Yes, this is multiple agents: advocates plus guardrails and judge to pressure-test options in parallel.

lane_b.framework_selector::_run_crew

Governance Sentinel

Risk and Controls

I challenge every recommendation against policy, risk, and observability before it reaches approval.

validators::*

Provenance Keeper

Audit Trail Owner

I lock every run into a verifiable record so decisions stay defensible in audits and board reviews.

provenance::ProvenanceLedger

Execution flow trace

Guided walkthrough of how control moves from query intake through CrewAI orchestration, validators, provenance, and learning loop.

1. Query intake

/generate receives intent + lane hint.
2. Intent routing

Classifier resolves lane and artifact type.
3. Master brain recall

Prior decisions + approvals are injected as precedent.
4. CrewAI stack bootstrap

Advocates, guardrails, and judge are wired.
5. Agent debate + scoring

Options debated, risk-reviewed, and ranked.
6. Validator spine

Quality, WAF, and schema checks run.
7. Provenance + learning loop

Hash-chain written and learning record persisted.

Why this trace matters

Shows exactly how a decision was produced, not just the final answer.
Makes governance visible with validator and HITL checkpoints in the control path.
Turns each run into reusable precedent through provenance and learning-loop memory.