Evidence-driven, Verifiable, Auditable Software Factory
EVA Foundry is an experimental AI-enabled software engineering ecosystem that implements the DPDCA loop (Discover → Plan → Do → Check → Act) for predictable, traceable, and auditable software delivery at scale.
EVA Foundry is not traditional "code vibes" software development. It's a data-driven software factory where:
- Every project phase is queryable via the Data Model API
- Every change has an audit trail through the Evidence Layer
- Requirements are always traceable with EVA Veritas
- AI agents query facts, not hallucinate by using the single source of truth
Traditional AI development: "Agent hallucinates route paths, auth logic, schemas → wastes turns fixing hallucinations"
EVA Factory: "Agent queries data model for EXACT route, auth, schemas, file location → zero hallucination"
graph TD
A[Human Intent: README → PLAN → STATUS] --> B[seed-from-plan.py]
B --> C[Data Model API<br/>37 layers, 27+ object types]
C --> D[Agent Skills<br/>Query model for context]
D --> E[DPDCA Execution Loop]
E --> F[Evidence Layer<br/>Immutable audit trail]
F --> G[Predictable Software Delivery]
| Project | Purpose | Status |
|---|---|---|
| 37-data-model | Single source of truth API (Cosmos DB + ACA) | 🟢 Production |
| 48-eva-veritas | Requirements traceability (CLI + MCP) | 🟢 Active |
| 40-eva-control-plane | Runtime evidence spine | 🟡 Active |
| 38-ado-poc | Scrum orchestration hub | 🟢 Active |
| 07-foundation-layer | Templates and patterns | 🟢 Active |
| Project | Purpose | Status |
|---|---|---|
| 31-eva-faces | Admin + Chat + Portal frontend (React) | 🟢 Production |
| 33-eva-brain-v2 | Agentic backend (FastAPI) | 🟢 Production |
| 44-eva-jp-spark | Bilingual GC AI assistant | 🟢 Phase 3 |
| 29-foundry | MCP servers + RAG + evaluation | 🟢 Active |
| Project | Purpose | Role |
|---|---|---|
| 97-workspace-notes | Workspace home root | Scripts, docs, Copilot config |
| 98-system-analysis | Azure subscription inventories | Reference (3 subs, 1,484+ resources) |
| 99-test-project | EVA Factory test vehicle | Script validation |
| 54-ai-engineering-hub | 93+ AI learning projects | Reference implementations |
Every sprint follows this deterministic pattern:
| Phase | Purpose | Key Operations |
|---|---|---|
| Discover | Synthesize state | Query WBS (undone stories), services (health), endpoints (implemented vs stubs), MTI audit |
| Plan | Select sprint stories | Filter WBS (status != done), size stories (XS/S/M/L), generate manifest, commit to .github/sprints/ |
| Do | Execute sprint | Agent reads data model, queries schemas, writes code with EVA-STORY tags, commits as feat(PROJECT-NN-NNN) |
| Check | Validate work | pytest (exit 0), veritas audit (MTI threshold), EVA-STORY coverage, data model consistency |
| Act | Record results | PUT story status=done, PUT endpoint status=implemented, reseed veritas-plan.json, POST /model/admin/commit |
29-foundry • 33-eva-brain-v2 • 34-eva-agents • 35-agentic-code-fixing • 54-ai-engineering-hub
31-eva-faces • 30-ui-bench • 43-spark • 44-eva-jp-spark • 45-aicoe-page
18-azure-best • 22-rg-sandbox • 28-rbac • 51-ACA • 98-system-analysis
01-documentation-generator • 09-eva-repo-documentation • 42-learn-foundry • 54-ai-engineering-hub
48-eva-veritas • 47-eva-mti • 49-eva-dtl
36-red-teaming • 99-test-project
38-ado-poc • 39-ado-dashboard • 40-eva-control-plane • 50-eva-ops
- Bootstrap your environment: Read 97-workspace-notes copilot-instructions
- Query the data model:
Invoke-RestMethod https://marco-eva-data-model.livelyflower-7990bc7b.canadacentral.azurecontainerapps.io/model/agent-summary - Check requirements traceability:
node 48-eva-veritas/src/cli.js audit-repo --repo ./your-project - Apply templates: Use 07-foundation-layer artifacts
# Step 1: Bootstrap the data model
$base = "https://marco-eva-data-model.livelyflower-7990bc7b.canadacentral.azurecontainerapps.io"
Invoke-RestMethod "$base/model/agent-guide"
# Step 2: Query project context
Invoke-RestMethod "$base/model/projects/{PROJECT_FOLDER}"
# Step 3: Get active WBS stories
Invoke-RestMethod "$base/model/wbs/?status=not-done"
# Step 4: Check endpoint schemas
Invoke-RestMethod "$base/model/endpoints/?status=implemented"Golden Rule: Query the data model FIRST. One HTTP call beats 10 file reads.
- Data Model Guide: 37-data-model/USER-GUIDE.md
- Azure Best Practices: 18-azure-best (32 entries)
- Workspace Rules: 97-workspace-notes/.github/copilot-instructions.md
- Data Model API: https://marco-eva-data-model.livelyflower-7990bc7b.canadacentral.azurecontainerapps.io
- Production Subscriptions: EsDAICoESub (1,250 resources) • EsPAICoESub (203 resources) • MarcoSub (31 resources)
EVA Foundry is an experimental research platform. All projects follow these principles:
- Evidence-First: Every recommendation backed by actual data
- Query-Driven: Data model is the single source of truth
- Honest by Default: No fluff, no marketing language, no overstating
- ASCII Only: Zero tolerance for Unicode/emojis in code and docs
- Traceable: EVA-STORY tags link code to requirements
Governance: See 97-workspace-notes for workspace-wide rules.
- 58 Public Repositories across all maturity levels (idea → poc → active → production → retired)
- 37 Layers in the data model (projects, services, endpoints, screens, containers, WBS, evidence, etc.)
- 1,484+ Azure Resources catalogued across 3 subscriptions
- 93+ AI Reference Projects in the learning hub
- Zero Hallucination Architecture via fact-based agent queries
Individual projects have their own licenses. See each repository for details.
- Organization: eva-foundry on GitHub
- Data Model API: https://marco-eva-data-model.livelyflower-7990bc7b.canadacentral.azurecontainerapps.io
- Workspace Root: 97-workspace-notes
Built with evidence. Verified with data. Auditable by design.