Secretary AI

An AI-powered Discord bot and meeting transcript integration system for business information management.

Features

• Meeting Transcript Processing: Automatically extract participants, projects, topics, tasks, and generate summaries from meeting transcripts using GPT-4.1-mini.
• Task Extraction: Detect and record explicitly assigned tasks with deadlines and assignees, linked back to meetings and members.
• Fuzzy Matching: Intelligently match extracted names to existing database records (supports full names, unique first names, and strips a bit of extra “noisy” text).
• File Watcher: Monitor a landing folder for new transcripts with an interactive renaming and routing flow.
• Discord Chatbot: Talk to Secretary AI in Discord to retrieve information (“What are my tasks?”, “Who is Sam Choong?”, “What did I miss?”) and perform allowed updates/creations.
• Agentic Tool Use: The chatbot decides which database tools to call, runs them, and answers based strictly on those results (no raw SQL from the model).
• Async Architecture: Built for performance with async database operations and API calls.

Quick Start

1. Install Dependencies

You can use either pip or uv. If you’re using uv, prefer running your app via uv run so it uses the right environment.

# Using pip + venv (recommended)
python -m venv .venv
.venv\Scripts\activate  # on Windows
pip install -r requirements.txt

# Or using uv
uv pip install -r requirements.txt

2. Configure Environment

Copy .env.example to .env and fill in your credentials:

cp .env.example .env

Required environment variables:

• DISCORD_TOKEN - Your Discord bot token
• OPENAI_API_KEY - Your OpenAI API key
• DATABASE_URL - PostgreSQL connection string

3. Run the Bot

# Start Discord bot
python main.py bot
# or with uv
uv run python main.py bot

# Start file watcher
python main.py watch
# or with uv
uv run python main.py watch

# Process a file directly
python main.py process landing/my_transcript.txt

Commands

CLI Commands

python main.py bot          # Start the Discord bot
python main.py watch        # Start the file watcher
python main.py process FILE # Process a specific transcript
python main.py setup        # Create/verify database tables
python main.py help         # Show help

Discord Chat Usage

• Mention the bot to chat: @SecretaryAI ...
• Example queries:
  • @SecretaryAI what are my current tasks?
  • @SecretaryAI what did I miss in the last full committee meeting?
  • @SecretaryAI who is Michael Huang?
  • @SecretaryAI create a new task to review recruitment email copy, due next Friday, and assign it to me.
  • @SecretaryAI add a project called "O-Week Preparation" and assign it to all executives.
  • @SecretaryAI what time is it right now?

Project Structure

secretary-ai/
├── main.py                    # Main entry point
├── discord_bot/
│   └── bot.py                 # Discord bot and chat logic
├── transcript_integrator/
│   ├── __init__.py
│   ├── integrator.py          # Main transcript processing engine
│   ├── models.py              # SQLAlchemy database models
│   ├── database_tools.py      # Async database tools used by the chatbot
│   └── file_watcher.py        # File monitoring and renaming
├── landing/                   # Drop transcript files here
│   ├── executive/
│   ├── projects_subcommittee/
│   ├── events_subcommittee/
│   ├── sponsorships_subcommittee/
│   ├── marketing_subcommittee/
│   ├── content-creation_subcommittee/
│   ├── hr_subcommittee/
│   ├── full/
│   └── unscheduled/
├── database-erd.txt           # Database schema reference
├── requirements.txt
├── pyproject.toml
└── .env                       # Environment configuration

Database Schema

The system uses the following tables:

Table	Description
committee	Organization members with Discord IDs and roles
meeting	Meeting records with summaries
meeting_members	Links meetings to attendees
meeting_projects	Links meetings to discussed projects
meeting_topics	Links meetings to discussion topics
meeting_tasks	Links meetings to assigned tasks
projects	Project information
project_members	Links projects to team members
tasks	Task records with deadlines
task_members	Links tasks to assignees
topic	Discussion topics

Meeting Types

The system supports the following meeting types:

• executive - Executive Committee Meeting
• projects_subcommittee - Projects Subcommittee Meeting
• events_subcommittee - Events Subcommittee Meeting
• sponsorships_subcommittee - Sponsorships Subcommittee Meeting
• marketing_subcommittee - Marketing Subcommittee Meeting
• content-creation_subcommittee - Content Creation Subcommittee Meeting
• hr_subcommittee - HR Subcommittee Meeting
• full - Full Committee Meeting
• unscheduled - Unscheduled / Ad-hoc Meeting

How It Works

File Watcher Flow

1. Drop a transcript file (.txt) into the landing/ folder
2. The watcher detects the new file and prompts you:
   • Select meeting type
   • Enter meeting date (DD-MM-YYYY)
   • Enter meeting name
   • Choose destination subfolder
3. File is renamed with INGESTED_ prefix and moved to the selected folder
4. Optionally run AI analysis to extract meeting information

Transcript Processing

1. Member Extraction: Identifies participants from the transcript and matches them to committee members using fuzzy matching over full names and first names.
2. Project Linking: Detects project mentions and links them to the meeting.
3. Topic Identification: Extracts discussion topics, linking to existing topics or creating new ones if needed.
4. Task Detection: Finds explicitly assigned tasks with deadlines and assignees (multiple assignees become multiple task_members rows).
5. Summary Generation: Creates a comprehensive meeting summary.

Fuzzy Matching

The system uses Python's difflib.get_close_matches() to handle:

• Typos and misspellings
• Name variations (plural/singular)
• Case differences

Default cutoffs:

• Members: 70% similarity
• Projects: 60% similarity
• Topics: 70% similarity

Configuration

Environment Variables

Variable	Required	Default	Description
DISCORD_TOKEN	Yes	-	Discord bot token
OPENAI_API_KEY	Yes	-	OpenAI API key
DATABASE_URL	Yes	-	PostgreSQL connection URL
OPENAI_MODEL	No	gpt-4.1-mini	OpenAI model to use
DISCORD_PROXY	No	-	Proxy for Discord connection

Example .env

DISCORD_TOKEN=your_discord_token_here
DATABASE_URL=postgresql://user:password@host:5432/database
OPENAI_API_KEY=sk-your-openai-key-here
OPENAI_MODEL=gpt-4.1-mini
# DISCORD_PROXY=socks5://127.0.0.1:7898

Development

Running Tests

pytest

Code Formatting

# Using black
black .

# Using ruff
ruff check --fix .

Troubleshooting

"No match found for member: X"

The member name in the transcript doesn't closely match any names in the committee table. Either:

• Add the member to the database
• Use a name that more closely matches existing records

"asyncpg.exceptions.UndefinedTableError"

The database tables don't exist. Run:

python main.py setup

"duplicate key value violates unique constraint ..._pkey"

If you see an error like:

duplicate key value violates unique constraint "tasks_pkey"
Key (task_id)=(1) already exists.

the PostgreSQL identity/sequence for that table is out of sync with existing data. You can fix it by resetting the sequence to MAX(id) + 1, for example:

SELECT setval(
  pg_get_serial_sequence('public.tasks', 'task_id'),
  COALESCE((SELECT MAX(task_id) FROM public.tasks), 0) + 1,
  false
);

Repeat with the appropriate table/column (e.g. public.topic / topic_id, public.meeting / meeting_id) if you see similar errors there.

Discord bot not responding

1. Check that DISCORD_TOKEN is set correctly
2. Ensure the bot has been invited to your server with proper permissions
3. Enable "Message Content Intent" in Discord Developer Portal
4. If behind a firewall, set DISCORD_PROXY

License

MIT License