Microsoft Fabric Projects: Five Portfolio Ideas for Job Seekers

The best Microsoft Fabric projects for a job seeker are small, end to end and easy to explain: data lands in OneLake, gets cleaned by a pipeline or notebook, is modelled in a semantic model and answers one business question in Power BI. Below are five projects, the Fabric items each one uses, where the data comes from, and what to show a recruiter or interviewer.

At a glance

Before you start: an environment that will not vanish

Most learners build on the Fabric trial. As of October 2026 it gives 60 days of an F4 or F64 trial capacity, up to 1 TB of OneLake storage and all the main workloads: Data Factory, Data Engineering, Data Science, Real-Time Intelligence and Power BI. Three limits shape your plan:

• Copilot, data agents, AI functions and AI services are not supported on trial capacity. Do not build your portfolio around them unless you have access to a paid capacity.
• When the trial ends, non-Power BI items become inactive. Content stays in OneLake for 7 days and can be reactivated only by moving the workspace to a paid capacity. Microsoft also says another trial is not assured.
• Fabric needs a work or school identity. If you only have Gmail, Microsoft documents a route to start a Fabric trial with a personal email by creating a Microsoft Entra ID user through an Azure account.

So treat the trial as a 60-day sprint. Connect the workspace to a GitHub repository on day one, record a short walkthrough video of each finished project, and keep the scripts that regenerate your data. Our guide to practising Microsoft Fabric without big bills covers the cost side in detail.

Project 1: Retail sales lakehouse with bronze, silver and gold layers

Business question: Which product categories and cities drive festive-season sales, and how do returns change the picture?

Data: Generate synthetic orders in a notebook with Indian cities and states, rupee amounts and a festive-season spike you design yourself, or use an open retail dataset whose licence clearly allows reuse. Synthetic data is fine, as long as you say so in the README; it also avoids any privacy question.

Fabric items:

• A lakehouse, with raw CSV files in Files/bronze and cleaned Delta tables in Tables.
• Notebooks (PySpark) for the silver layer (types, duplicates, bad rows) and the gold layer (a star schema with a sales fact and date, product and city dimensions).
• A pipeline that runs the notebooks in order on a schedule.
• A semantic model in Direct Lake mode and a Power BI report on top.

A silver-layer cell can be this short:

from pyspark.sql import functions as F

raw = spark.read.option("header", True).csv("Files/bronze/orders/*.csv")
silver = (raw
    .withColumn("order_date", F.to_date("order_date"))
    .withColumn("amount_inr", F.col("amount_inr").cast("decimal(12,2)"))
    .dropDuplicates(["order_id"])
    .filter(F.col("amount_inr") > 0))
silver.write.mode("overwrite").format("delta").saveAsTable("silver_orders")

What to show: a one-page architecture diagram, row counts per layer, the rules you used to reject rows, and two or three DAX measures you can explain line by line. If the layered design is new to you, read our explainer on medallion architecture in a Fabric lakehouse first.

Project 2: Incremental loads from a public API

Business question: Can the pipeline pick up only new records each day, without duplicates, and tell you when it fails?

Data: Any public REST API that returns records with an ID or a timestamp, for example a public weather, transport or open-data API. Read its terms of use and respect its rate limits.

Fabric items:

• A pipeline with a Copy activity from the REST source into a staging table, driven by parameters for the start date.
• A small watermark table that stores the last loaded value.
• A notebook that merges staged rows into the silver table.

-- Spark SQL in a Fabric notebook: upsert staged rows into silver
MERGE INTO silver_readings AS t
USING staged_readings AS s
  ON t.reading_id = s.reading_id
WHEN MATCHED THEN UPDATE SET *
WHEN NOT MATCHED THEN INSERT *;

What to show: two runs side by side, where the second run copies only new rows; how you handle late or duplicate records; and an alert on failure. These map directly to lines in the DP-700 study guide, such as designing full and incremental loads and handling duplicate, missing and late-arriving data.

Project 3: A real-time monitor with Eventstream and KQL

Business question: Where is a shared-bike or bus service under pressure right now, and who gets alerted?

Data: Eventstream includes built-in sample data sources: Bicycles, Yellow Taxi, Stock Market, Buses and others. They stream continuously, so you can build a live project without any device or paid service.

Fabric items: an eventstream that filters and routes events, an eventhouse with a KQL database, a KQL queryset, a Real-Time Dashboard and an Activator rule that fires when, for example, a docking point has no empty docks.

// Average bikes available per neighbourhood, last hour, 5-minute bins
BikeEvents
| where ingestion_time() > ago(1h)
| summarize AvgBikes = avg(No_Bikes) by Neighbourhood, bin(ingestion_time(), 5m)
| order by Neighbourhood asc

What to show: the dashboard refreshing in a screen recording, the KQL behind each tile, and a sentence on why you chose an eventhouse over a lakehouse for this data. Our article on Real-Time Intelligence in Fabric explains the pieces.

Project 4: From an app database to a report

Business question: A small shop or college library records transactions in a database. What do the owners need to see each month?

Data: Create your own tables (customers, orders, books, loans) and insert realistic sample rows with a script.

Fabric items: a SQL database in Fabric. Microsoft's SQL database overview explains that its data is replicated automatically into OneLake and exposed through a SQL analytics endpoint, so you can analyse it without building an ETL pipeline. A trial capacity allows up to three SQL databases. Add views on the analytics endpoint, a semantic model and a report.

CREATE VIEW dbo.vw_monthly_sales AS
SELECT
    DATEFROMPARTS(YEAR(o.order_date), MONTH(o.order_date), 1) AS month_start,
    c.city,
    SUM(o.amount_inr) AS sales_inr,
    COUNT(*) AS orders
FROM dbo.orders AS o
JOIN dbo.customers AS c ON c.customer_id = o.customer_id
GROUP BY DATEFROMPARTS(YEAR(o.order_date), MONTH(o.order_date), 1), c.city;

What to show: your table design, the star schema you derived from it, and a short note on why the transactional tables and the reporting views are kept separate. This covers DP-600 skills such as creating views and implementing a star schema.

Project 5: Make any project production-shaped

Interviewers often ask how you would work in a team. This project answers that by upgrading one of the four above.

• Git integration. Connect the workspace to GitHub. Fabric Git integration supports Azure DevOps, GitHub and GitHub Enterprise, and versions items such as lakehouses, notebooks, pipelines, semantic models and reports. Note that table data and files in the lakehouse are not stored in Git.
• Deployment pipelines. Promote changes from a development workspace to a test workspace.
• Security. Add row-level security to the semantic model so a "city manager" role sees only one city, and test the role before you share anything.

What to show: the repository with a clean commit history, a README that explains how to deploy the project into a fresh workspace, and a screenshot of the security role working. Version control, deployment pipelines and row-level security all appear in both the DP-600 study guide and the DP-700 guide.

What to put in front of recruiters

A recruiter skims a portfolio; an interviewer digs into it. Build for both.

• Lead with the question and the answer. The first lines of each README should say what business question the project answers and what you found.
• One diagram per project. Boxes for sources, Fabric items and the report, with arrows. Nothing fancier is needed.
• A two-minute video. Trial workspaces expire, so a screen recording is the only demo that lasts. Upload it to YouTube as unlisted and link it from the README.
• Your own measured numbers. Row counts, refresh duration, rows rejected. Never invent business results.
• Be careful with Publish to web. Microsoft warns that anyone on the internet can view a report published this way, including the underlying data, and it does not support reports that use row-level security. Use it only with synthetic or public data, or share screenshots instead.
• Map each project to the exam. List the DP-600 or DP-700 skills each project covers. It shows intent and gives the interviewer an easy set of questions to ask you.

How CertTulen Academy can help

CertTulen Academy runs live, MCT-led online classes with hands-on Fabric labs, which give you a guided first pass at the same items these projects use. Microsoft Fabric Analytics Engineer (DP-600) suits analysts who want to own lakehouses, warehouses and semantic models; Microsoft Fabric Data Engineer (DP-700) suits those heading for pipelines, Spark and streaming. Each course page lists what is included, including lab access. Contact us or WhatsApp +91 93618 83583 for a batch that suits your time zone and schedule.