Medallion Architecture

**On screen:** Bronze layer node — Raw Data, source format preserved, append-only history - This is the landing zone for everything coming into the platform - Data keeps its original shape — CSV, JSON, database tables, API payloads - Append-only means we never delete or overwrite; full history is preserved - Think of it as the "safety net" — if anything goes wrong downstream, you can always come back here *Transition:* Arrow labeled "Clean" points to Silver — that's the next transformation step

**On screen:** Silver layer node — Clean Data, validated, de-duplicated, conformed types - This is where the heavy lifting happens: cleaning, deduplication, type conformance - Business rules start being applied — nulls handled, dates standardized, keys aligned - The data is now trustworthy enough for analysts and data scientists to query directly - Still not shaped for reporting — that's Gold's job *Transition:* Arrow labeled "Model" points to Gold — the final curation step

**On screen:** Gold layer node — Curated Data, star schema modeled, optimized for reporting - Star schema: fact tables surrounded by dimension tables - Pre-computed aggregations and KPIs live here - This is the layer Power BI, Excel, and other reporting tools should connect to - Optimized for read performance — queries are fast because the modeling work is already done *Key message:* Data flows Bronze to Silver to Gold through transformation steps between each layer

**On screen:** Bronze Layer frame appears — fieldset with orange border - We're now looking at who actually uses each layer - Bronze isn't just a dumping ground — specific roles depend on it - The frame groups personas on the left and use cases on the right *Transition:* Personas appear next

**On screen:** Bronze Personas — Data Engineers, Platform Admins, DevOps Engineers - Data Engineers build and maintain the ingestion pipelines that feed Bronze - Platform Admins manage the infrastructure — storage accounts, access policies - DevOps Engineers handle CI/CD for the data pipelines themselves - These are all technical roles — Bronze is engineering territory

**On screen:** Bronze Use Cases — Audit & Compliance, Data Lineage, Source Recovery - Audit teams need the unaltered source record for regulatory compliance - Data Lineage traces where every value originated — Bronze is the starting point - Source Recovery: if a downstream transform goes wrong, Bronze is your rollback *Transition:* Moving to the Silver layer and its audience

**On screen:** Silver Layer frame appears — fieldset with silver/gray border - Silver serves a different audience: people who need clean, queryable data - Not yet modeled for reporting, but trustworthy enough for exploration and ML *Transition:* Personas appear next

**On screen:** Silver Personas — Data Scientists, Data Analysts, ML Engineers - Data Scientists explore Silver for feature engineering and model training - Data Analysts run ad-hoc queries — Silver is clean enough for reliable answers - ML Engineers build training pipelines that pull directly from Silver tables

**On screen:** Silver Use Cases — Feature Engineering, Exploratory Analysis, Data Quality Validation - Feature Engineering: building ML features from cleaned, conformed data - Exploratory Analysis: ad-hoc querying to find patterns before formal reporting - Data Quality Validation: Silver is where you verify cleaning rules actually worked *Transition:* Moving to the Gold layer — the business consumption tier

**On screen:** Gold Layer frame appears — fieldset with gold border - Gold is the business-facing layer — everything here is optimized for consumption - This is where the non-technical audience enters the picture *Transition:* Personas appear next

**On screen:** Gold Personas — Report Consumers, Self-service Analysts, Business Leaders - Report Consumers view dashboards and subscribe to scheduled reports - Self-service Analysts build their own Power BI reports and Excel analyses - Business Leaders make decisions based on KPIs and executive dashboards

**On screen:** Gold Use Cases — Power BI Dashboards, AI / Copilot, Excel & Paginated - Power BI Dashboards: interactive visuals connected to the semantic model - AI / Copilot: natural language queries against curated, well-modeled data - Excel & Paginated: traditional reporting formats, pulling from the same Gold layer *Key message:* The layer determines the audience — raw for engineering, clean for analysis, curated for business

**On screen:** Data Engineering pillar card — Bronze (ingest raw), Silver (clean & validate), Gold (model & curate) - Data Engineering is the *process*, not a product - Any tool that moves data through Bronze, Silver, and Gold stages is doing data engineering - Callout reinforces: "This is the process, not a product" - Fabric uses Dataflows, Notebooks, and Pipelines — but the concept is tool-agnostic *Transition:* Next pillar shows what sits between engineering and consumption

**On screen:** Semantic Layer pillar card — Semantic Model, SQL Endpoint, KQL Queryset - These are the *products* — the connection points between data and consumers - Semantic Model is the primary path: measures, relationships, DAX logic - SQL Endpoint provides T-SQL access for notebooks and ad-hoc queries - KQL Queryset enables real-time analytics scenarios - Callout: "These are connection inputs — the bridge between data engineering output and business consumption" *Transition:* Final pillar shows who actually consumes the data

**On screen:** Reporting and BI pillar card — Power BI, Excel, Paginated Reports, AI / Copilot - These are the *consumers* — the tools end users actually interact with - Power BI for interactive dashboards, Excel for spreadsheet analysis - Paginated Reports for pixel-perfect, printable output - AI / Copilot for natural language querying - Callout: "These tools read from the semantic layer — they don't touch Bronze/Silver/Gold directly" *Key message:* Three pillars: the process (engineering), the products (semantic), the consumers (reporting)

**On screen:** Bronze feature card — "Iron Ore" subtitle; Format: Source, Schema: Source system, Write: Append-only, History: Full - The metal refining analogy: Bronze is like iron ore — raw, unprocessed - Format preserves whatever the source system sent — CSV, JSON, database dumps - Schema matches the source system exactly — no transformation yet - Append-only write pattern means nothing is ever overwritten or deleted - Full history is the key value: complete audit trail from day one *Transition:* Silver card shows what happens after the first refinement pass

**On screen:** Silver feature card — "Forged Metal" subtitle; Format: Delta, Schema: Conformed, Write: Versioned, Quality: ACID - Forged Metal analogy: raw material has been refined into something useful - Delta format = Parquet files + transaction log — enables versioning and time travel - Schema is now conformed: consistent data types, naming conventions, key alignment - Versioned writes mean you can roll back to any previous state - ACID guarantees: Atomicity, Consistency, Isolation, Durability *Transition:* Gold card reveals the final refinement stage

**On screen:** Gold feature card — "Refined Product" subtitle; Format: Delta, Schema: Star schema, Content: Facts + dims, KPIs: Pre-computed - Refined Product analogy: the finished goods ready for the consumer - Still Delta format, but with enforced star schema on top - Star schema: fact tables (transactions, events) surrounded by dimension tables (products, dates, customers) - Pre-computed KPIs: aggregations calculated in advance so reports load fast *Key message:* Each layer maps to a real file format progression — from raw source files through Delta tables to modeled star schemas

**On screen:** Core medallion layers appear — Raw (Bronze), Standardized (Silver), Curated (Gold) with arrows between them - The intro text sets context: core medallion is Bronze, Silver, Gold - These three cards show the standard pattern that every organization uses - Raw = source format preserved, Standardized = Delta with validation, Curated = star schema with business logic - Arrows show the left-to-right transformation flow *Transition:* Extension layers will appear on either side

**On screen:** Landing card appears to the left of the core layers — "Staging" subtitle, dashed extension border - Landing is a pre-Bronze extension: an exact, unchanged copy from the source system - Same format, same schema — no transformation at all - Purpose: decouple ingestion from processing so source systems aren't queried repeatedly - The dashed border and legend distinguish extensions from core layers *Transition:* Semantic extension appears on the right side

**On screen:** Semantic card appears to the right of Gold — "Platinum" subtitle, dashed extension border - Semantic / Platinum is a post-Gold extension: TMDL metadata layered on top of curated data - Measures, relationships, hierarchies, and business-friendly names defined here - This is where the semantic model lives — the bridge to reporting tools - Some organizations fold this into Gold; others keep it separate for clarity *Transition:* Info box summarizes the pattern

**On screen:** Info box with two bullet points about Semantic layer flexibility and naming conventions - First point: the Semantic layer can be its own step or fold into Gold — it depends on org preference - Second point: Bronze, Silver, Gold are conventions, not standards — organizations often rename them - This is a good moment to ask the audience what their org calls these layers *Key message:* Core three are the standard medallion; Landing and Semantic are common extensions that add granularity

**On screen:** Sources zone — ERP, CRM, and Files icons in the leftmost column - This is the full end-to-end architecture diagram — everything we've discussed comes together here - ERP systems (SAP, Dynamics), CRM platforms (Salesforce, HubSpot), flat files (CSV, Excel, JSON) - These are the origin points — data flows from left to right across the diagram - Every organization has multiple source systems; the architecture handles them all the same way *Transition:* Data Engineering zone shows how source data gets processed

**On screen:** Data Engineering zone — three sub-layers: Ingest (Data Factory), Transform (Prepare + Curate), Store (Bronze, Silver, Gold) - Ingest Layer: Data Factory pulls from source systems on a schedule or trigger - Transform Layer has two stages: Prepare (cleanse + validate) then Curate (warehouse build) - Store Layer maps directly to the medallion: Bronze (raw), Silver (clean), Gold (curated) - This is the most complex zone — it's where all the data engineering work happens - Flow overlay arrows will connect these nodes to downstream consumers *Transition:* Semantic Layer shows the access points

**On screen:** Semantic Layer zone — Semantic Model (primary, filled) and SQL Endpoint (bordered) - Two access paths to the data, each serving different consumers - Semantic Model is the primary path: DAX measures, relationships, hierarchies for Power BI - SQL Endpoint provides T-SQL access for notebooks, ad-hoc queries, and cross-platform tools - The visual weight (filled vs bordered) signals which path is primary *Transition:* Reporting zone shows the consumer groups

**On screen:** Reporting and BI zone — three sub-layers: Data Science (Semantic Link + Notebooks), Insights (AI / Copilot), Reporting (Reports, Excel, Paginated) - Data Science: Semantic Link connects notebooks to the semantic model; multiple notebooks shown for parallel workloads - Insights: AI / Copilot provides natural language querying against the curated data - Reporting: Power BI reports, Excel workbooks, and Paginated reports for traditional BI - All three sub-layers read from the Semantic Layer — none bypass it to hit raw tables *Transition:* Governance wraps around the entire pipeline

**On screen:** Governance & Lifecycle bar — Plan, Implement, Monitor, Govern, Secure, Deploy - Governance isn't a separate step — it wraps the entire pipeline from end to end - Plan: define data policies, ownership, and quality standards before building - Implement: build the pipelines, models, and reports following those standards - Monitor: track data freshness, pipeline failures, and usage patterns - Govern: enforce access controls, data classification, and retention policies - Secure: encrypt at rest and in transit, manage identity and credentials - Deploy: CI/CD for data assets — version-controlled promotions across environments *Key message:* The full picture ties together sources, engineering, semantic, reporting, and governance into one coherent architecture