AI Meeting Notes Automation

This project automates the extraction, summarization, and documentation of Microsoft Teams client advisory calls. Built using Azure services, Microsoft Graph API, and OpenAI, the solution captures Outlook meeting metadata, retrieves Teams transcripts, summarizes them using GPT-4, stores data in Azure SQL, and updates CRM (ClickUp) all on a scheduled basis.

Tech Stack

• Microsoft Graph API – Fetches Outlook calendar and Teams transcript data.
• Azure Function App – Automates daily script execution.
• Azure OpenAI (GPT-4) – Summarizes meeting transcripts.
• Azure SQL Database – Stores meeting and summary data.
• ClickUp API – Updates CRM with summarized notes.
• Python – Primary programming language for all logic.

Architecture

List view of MS Teams Meetings in ClickUp

AI Output in its respective ClickUp task

Project Structure

.
├── function_app.py          # Main script triggered by Azure Function
├── api_graph.py             # Microsoft Graph API interaction logic
├── api_clickup.py           # ClickUp API integration (create/update tasks)
├── api_gpt.py               # Azure OpenAI (GPT-4) integration for summarization
├── data_process.py          # Meeting ID comparison and event data parsing
├── sql_azure_db.py          # Azure SQL DB interaction (insert/update records)
├── utils.py                 # Logging utilities
├── requirements.txt         # Python dependencies
└── README.md                # Project documentation

function_app.py

• Main Azure Function trigger
• Orchestrates the full pipeline from data fetch → summarization → ClickUp update

api_graph.py

• Handles Microsoft Graph authentication and API calls:
  • Get calendar events
  • Retrieve Teams meeting transcripts
  • Clean and extract VTT text
  • Look up user ID by email

api_clickup.py

• Handles communication with ClickUp API:
  • Create or update ClickUp tasks
  • Attach summaries and metadata
  • Automate task status updates

api_gpt.py

• Connects to Azure OpenAI (GPT-4) for:
  • Summarizing raw transcript text
  • Ensuring concise and accurate summaries for documentation

data_process.py

• Extracts meeting IDs and compares them
• Formats event metadata
• Computes durations and attendee lists

sql_azure_db.py

• Connects to Azure SQL Server
• Inserts and updates records in the tblOutlookEventsY table

utils.py

• Logging utilities for tracking process flow and errors

Workflow Overview

1. Trigger
   Azure Function App triggers function_app.py on a schedule (e.g. daily at 7 PM SGT).

2. Fetch Calendar Events
   api_graph.get_outlook_metadata() pulls relevant meetings from Outlook Calendar within the defined date range.

3. Process Event Metadata
   data_process.parse_event() extracts and cleans relevant data fields.

4. Get Meeting Transcript
   api_graph.get_transcript_content_url() fetches the transcript URL for Teams calls. Then, api_graph.get_filtered_vtt() cleans the transcript (from VTT format).

5. Summarize Using OpenAI GPT-4
   Transcript text is sent to Azure OpenAI, which returns a structured summary.

6. Update Azure SQL Database
   sql_azure_db.sql_insert_new_record() or sql_update_record() inserts or updates meeting info and summaries.

7. Update CRM (ClickUp)
   The summary is pushed to the relevant ClickUp task using the ClickUp API.

Key Features

• Transcript Matching
  Uses encoded vs. raw meeting ID comparison for precise transcript mapping.

• Resilient Logging
  All critical steps are logged using Python's logging module via utils.py.

• Extensible Database
  SQL table tblOutlookEventsY tracks each meeting’s metadata, processing status, and summary content.

• CRM Integration
  (Optional) Each meeting can be linked to a ClickUp task for automated note updates.

Deployment

All Python files are deployed inside an Azure Function App, which runs automatically on a timer trigger. Environment variables (e.g., client credentials, SQL settings) are stored securely in the Azure Function configuration settings.

Environment Variables (for local testing)

Create an .env file or set these as App Settings in your Azure Function App:

GRAPH_APP_URL=
GRAPH_APP_CLIENT_ID=
GRAPH_APP_CLIENT_SECRET=

SQL_SERVER=
SQL_USERNAME=
SQL_PASSWORD=
SQL_DATABASE=
SQL_DRIVER={ODBC Driver 18 for SQL Server}

CLICKUP_API_TOKEN= 
CLICKUP_USERS_LIST_ID= 

OPENAI_API_KEY= 
OPENAI_URL= 

Requirements

pip install -r requirements.txt

Make sure to include pyodbc, requests and pytz

🔍 Sample SQL Table Structure (tblOutlookEventsY)

CREATE TABLE tblOutlookEventsY (
    event_id NVARCHAR(255) PRIMARY KEY,
    joinURL_id NVARCHAR(MAX),
    is_cancelled BIT,
    is_organizer BIT,
    event_type NVARCHAR(100),
    is_online_meeting BIT,
    online_meeting_provider NVARCHAR(100),
    response_status NVARCHAR(100),
    subject NVARCHAR(500),
    organizer NVARCHAR(255),
    start_time DATETIME,
    end_time DATETIME,
    location NVARCHAR(255),
    categories NVARCHAR(MAX),
    duration NVARCHAR(20),
    attendees NVARCHAR(MAX),
    is_recording_exist BIT,
    is_transcript_exist BIT,
    get_transcript BIT,
    get_transcript_done BIT,
    summarize_transcript_done BIT,
    summarized_transcript NVARCHAR(MAX),
    clickup_task_id NVARCHAR(100),
    update_clickup_done BIT
);

Testing Locally

You can test the main script by running:

python function_app.py

Ensure your environment variables are loaded or defined before running.

Potential Extensions

Scalable Diagnostic Intelligence - As it is designed with scalability in mind, we can generate diagnostics based on previous meeting summaries. The system can be enhanced to automatically assess whether a client advisory session requires further action or indicates resolution/no action needed. Sentiment & Intent Analysis - Layer in sentiment analysis and intent classification using Azure Cognitive Services to flag urgent or sensitive client conversations. CRM Integration Flexibility - Expand support for additional CRMs beyond ClickUp via modular API adapters. Smart Filtering & Alerts - Auto-prioritize meetings for follow-up based on keyword triggers or customer profile metadata. Smart Filtering & Alerts - Auto-prioritize meetings for follow-up based on keyword triggers or customer profile metadata. UI Dashboard (Ongoing Enhancement) - Create a frontend dashboard for manual review, editing, and audit trail of summaries and automation status.

License

This project is internal-use. Please customize depending on your organization’s policies.

Author

Built by Deodie Picson as part of a productivity and efficiency automation initiative. For inquiries or support, please contact deodie.dev@gmail.com.