📦 Series: Exam Auto-Evaluator Part 1 of 3 — Concepts & Architecture

"Before you build anything, you need to see the complete picture. Not just what each tool does — but exactly how data flows from a student's upload to an approved answer sheet. This post gives you that picture."

In this series you build a real enterprise automation: a student uploads an Excel file full of exam questions to a chat agent. The agent automatically answers every question using AI, saves the results to SharePoint, and sends you an approval email before the answers go back to the student. No manual work. No reviewing one question at a time.

This post explains what each tool is, why it exists, and — most importantly — exactly what data moves between them. By the end you will be able to draw the complete architecture from memory.

The Complete Picture — Before Reading Anything Else

Here is the full system you are building. Every concept in this post maps to one of these boxes.

┌─────────────────────────────────────────────────────────────────────┐
│                    EXAM AUTO-EVALUATOR PIPELINE                     │
└─────────────────────────────────────────────────────────────────────┘

STEP 1          STEP 2              STEP 3
Student         Copilot Studio      Power Automate
types message   agent responds,     Flow starts,
in chat    →    asks for file   →   saves file to
                                    OneDrive

                                         ↓

STEP 6          STEP 5              STEP 4
Student         If approved →       AI Builder reads
notified        email sent      ←   each question,
✅              to student          writes answers,
                                    saves to SharePoint

                    ↓
              STEP 4b
              Approval email
              sent to Raushan
              for review

─────────────────────────────────────────────────────────────────────

DATA THAT MOVES AT EACH STEP:

Step 1→2:  Student types "evaluate exam" in chat
Step 2→3:  Student uploads exam-questions.xlsx
           Copilot Studio saves file to OneDrive automatically
           Then triggers the Power Automate flow (no file passed — path passed)
Step 3→4:  Power Automate reads rows from Excel in OneDrive
           Sends each question to AI Builder
           AI Builder returns answer for each question
Step 4→5:  All Q+A pairs written to new Excel in SharePoint
           SharePoint link generated
           Approval email sent with the link
Step 5→6:  Reviewer clicks Approve in email
           Power Automate detects approval
           Sends confirmation to student
    

Tool 1 — Microsoft Copilot Studio

What problem existed before it

Building a conversational agent used to require custom code, expensive third-party platforms, or months of development. Business users could not build agents without engineering help. Microsoft built Copilot Studio to close this gap.

What it is — in one sentence

Copilot Studio is the front door of your automation. It is what the student talks to. It understands what the student wants, asks for the file, receives it, saves it to OneDrive, and then triggers Power Automate to do the heavy work.

🎯 Real-world analogy Copilot Studio is like a receptionist. The receptionist greets you, takes your documents, puts them in the right tray, and calls the back office to say "new documents arrived, please process." The receptionist does not process anything — they coordinate.

The four things you build inside Copilot Studio

Thing you build
What it is
In our project
Agent
The AI assistant the student chats with
Named "Exam Evaluator"
Topic
One complete conversation flow — a trigger phrase + steps that follow
Named "Evaluate Exam Paper" — triggered when student types "evaluate exam"
Question node
A step that asks the user for input and stores their response in a variable
Asks for the Excel file, stores it in a variable called ExamFile
Action node
A step that calls an external flow or service
Calls the Power Automate flow and passes the OneDrive file path

What Copilot Studio does NOT do

Copilot Studio does not read Excel files. It does not call AI directly. It does not send emails. It does not save to SharePoint. All of that is Power Automate's job. Copilot Studio only handles the conversation and passes control to Power Automate at the right moment.

Tool 2 — Microsoft Power Automate

What problem existed before it

Connecting services — reading a file, sending an email, saving to SharePoint, waiting for approval — required custom code or expensive integration tools. Power Automate lets anyone build these workflows visually.

What it is — in one sentence

Power Automate is the back office that processes everything after Copilot Studio hands off. It reads the Excel file, calls AI Builder for each question, writes the answers to SharePoint, and sends the approval email.

🎯 Real-world analogy Power Automate is an assembly line. Station 1 reads the incoming file. Station 2 sends each question to the AI. Station 3 writes answers to a new file. Station 4 saves the file and sends email. Each station does one job and passes the result to the next.

The specific flow type we use — Agent Flow

Power Automate has three flow types. We use Agent Flow.

Flow type
How it starts
Use for
Automated Cloud Flow
An event (new email, new file, form submit)
Connecting cloud services automatically
Scheduled Cloud Flow
A timer (every day at 9am)
Regular batch processing
Agent Flow ← THIS ONE
Copilot Studio calls it
Processing triggered by a chat agent

An Agent Flow has two special nodes pre-built when you create it: "When an agent calls the flow" at the top and "Respond to the agent" at the bottom. You build your steps between them. Never delete these two nodes.

Tool 3 — AI Builder

What problem existed before it

Adding AI to a Power Automate flow meant calling Azure OpenAI APIs directly — which required an Azure subscription, API keys, JSON knowledge, and custom code. AI Builder wraps this into a simple action anyone can use.

What it is — in one sentence

AI Builder is a ready-to-use AI assistant sitting inside Power Automate. You write a prompt describing what the AI should do, pass it a question, and it returns the answer as plain text you can use in the next step.

🎯 Real-world analogy AI Builder is like having a knowledgeable colleague available inside your assembly line. You hand them one question at a time. They answer it. You take the answer and write it into the output file. They never see the whole file — just one question at a time.

The action we use: "Run a prompt"

Inside Power Automate, search for "Run a prompt". This is the AI Builder action for text generation. Older tutorials may call it "Create text with GPT" — same feature, renamed in May 2025.

Licence requirement

Testing AI Builder prompts in the Prompt Builder editor is completely free. Running the action inside a deployed flow requires AI Builder credits — included in the Power Automate Premium 90-day trial.

💡 Azure OpenAI alternative If your organisation already has Azure OpenAI, you can replace AI Builder with an HTTP action calling the Chat Completions API directly. This gives full control over model, temperature, and token limits. We cover the AI Builder approach here as it requires no Azure subscription.

How the Three Tools Connect — Data Flow Diagram

┌──────────────────────────────────────────────────────────────────┐
│                    DATA FLOW BETWEEN TOOLS                       │
└──────────────────────────────────────────────────────────────────┘

COPILOT STUDIO                    What data moves
──────────────────────────────────────────────────
Student chats → agent receives    text message
Agent asks for file               text response
Student uploads Excel             Excel file (binary)
Copilot Studio saves to OneDrive  file saved to /ExamProcessing/
Copilot Studio calls flow    →→→  passes: OneDrive file path (text)
                                           + timestamp (text)

POWER AUTOMATE (Agent Flow)       What data moves
──────────────────────────────────────────────────
Receives file path from Copilot   text: "/ExamProcessing/exam.xlsx"
Reads Excel rows (List rows)      array: [{Question: "..."}, ...]
Loop: for each row           →→→  sends: one question (text) to AI Builder

AI BUILDER (Run a prompt)         What data moves
──────────────────────────────────────────────────
Receives question text            text: "What is CALCULATE in DAX?"
Returns answer text          ←←←  text: "CALCULATE modifies filter context..."

POWER AUTOMATE (continues)        What data moves
──────────────────────────────────────────────────
Writes Q+A to output Excel        rows added to new Excel file
Saves Excel to SharePoint         file saved, URL retrieved
Sends approval email              email with SharePoint link
Waits for approval response       Approve or Reject (text)
If Approved → emails student      confirmation email sent

COPILOT STUDIO (receives result)  What data moves
──────────────────────────────────────────────────
Flow returns status text     ←←←  text: "Processing complete"
Agent sends message to student    "Your exam is being reviewed"
    

Prerequisites — Confirm All of These Before Part 2

1. Microsoft 365 work account

You need a work or school Microsoft 365 account in the format yourname@yourorganisation.com. Personal accounts (gmail.com, outlook.com, hotmail.com) do not have access to Copilot Studio trial environments. If you only have a personal account, sign up for a Microsoft 365 Business Basic trial at microsoft.com/en-us/microsoft-365/business/compare-all-plans.

2. Copilot Studio trial — exact steps

  1. Open a browser and go to copilotstudio.microsoft.com
  2. Click Sign in — use your work account (yourname@yourorganisation.com)
  3. If the page shows "You need a licence" or "Start trial", click Try free or Start trial
  4. Select your country → click Get started
  5. A spinner shows "Setting up your environment" — wait 1-3 minutes
  6. You land on the Copilot Studio home page showing a left sidebar with: Home, Agents, Flows, Tools. You are ready.

3. Power Automate Premium trial — exact steps

  1. Go to powerautomate.microsoft.com
  2. Sign in with the same work account
  3. Click your profile picture (top right) → look for Try premium or go to powerautomate.microsoft.com/pricing → click Try free under Premium
  4. Confirm the 90-day trial → click Start trial
  5. Trial activates immediately. You now have access to AI Builder and premium connectors.

4. Create your sample Excel file — exact steps

  1. Open excel.office.com or Excel desktop
  2. Create a new blank workbook
  3. In cell A1 type: Question
  4. In cell B1 type: Answer
  5. Add questions in column A starting from row 2:
    • A2: What is CALCULATE in DAX?
    • A3: Explain Star Schema in Power BI
    • A4: What is the difference between a measure and a calculated column?
  6. Leave column B completely empty — AI will fill these
  7. Select cells A1 to B4
  8. Click Insert in the top ribbon → click Table
  9. A dialog appears: confirm the range is correct → make sure "My table has headers" is checked → click OK
  10. The data turns into a styled Excel Table
  11. Name the table: click anywhere inside the table → the Table Design tab appears in the ribbon → on the far left find the Table Name field (showing "Table1") → click it → type ExamQuestions → press Enter
  12. Save the file: click File → Save As → OneDrive - [Your Organisation Name] (not personal OneDrive — must be your work OneDrive)
  13. Create a folder called ExamProcessing and save the file there as exam-questions.xlsx
⚠️ OneDrive for Business vs Personal Save to OneDrive for Business — the one showing your organisation name. If you see "OneDrive - Personal" that is the wrong one. Power Automate's Excel connector reads from OneDrive for Business only.

5. SharePoint site

You need access to any SharePoint site in your Microsoft 365 tenant with a document library. To check: go to sharepoint.com → sign in → you should see at least one site listed. If you see none, ask your IT admin to create a team site for you, or create one via Microsoft Teams → create a new Team → it automatically creates a SharePoint site.

🎯 Quick Check — Before Moving to Part 2

Q1: Which tool handles the student-facing conversation? Which tool reads the Excel file?

Show Answer

Copilot Studio handles the conversation — it is what the student chats with. Power Automate reads the Excel file — Copilot Studio passes the file path to Power Automate, which then reads the rows. Copilot Studio never reads the Excel file directly.

Q2: What does Copilot Studio pass to Power Automate — the file itself or the file path?

Show Answer

The file path (text). In this architecture, Copilot Studio saves the uploaded file to a known OneDrive folder first. Then it passes the file path as a text string to Power Automate. Power Automate reads the file from OneDrive using that path. This avoids file type compatibility issues between the two platforms.

Q3: What is the difference between "Automated Cloud Flow" and "Agent Flow" in Power Automate?

Show Answer

An Automated Cloud Flow starts on its own when an event happens (new email, new SharePoint file, form submitted). An Agent Flow starts when Copilot Studio calls it — it has a special trigger "When an agent calls the flow" and a special response "Respond to the agent". You must create an Agent Flow (not a Cloud Flow) for this project — only Agent Flows appear as callable actions inside Copilot Studio topics.

Key Takeaways — Part 1

  • Copilot Studio = front door (conversation, file upload, triggers the flow)
  • Power Automate = back office (reads Excel, calls AI, saves to SharePoint, sends email)
  • AI Builder = the AI assistant inside Power Automate ("Run a prompt" action)
  • Data flow: student uploads file → saved to OneDrive → path passed to flow → AI answers questions → saved to SharePoint → approval email sent
  • You need: M365 work account + Copilot Studio trial + Power Automate Premium trial + Excel file saved to OneDrive for Business