Current AI

The zero-friction AI Project Management tool for Discord. It turns your natural chat into organized, actionable tasks.

Comment

Current-AI

Inspiration

After attending over 21 hackathons, I've witnessed the same problem again and again: teams spend more time organizing than building.

Picture this: It's 2 AM at a hackathon. Your team is exhausted. Someone asks, "Who's working on the database?" Silence. Another person chimes in, "I thought you were doing that? I've been building the API!" Now you have duplicate work, wasted hours, and mounting frustration. Sound familiar?

Hackathon Confusion

The reality is brutal:

  • 🔴 6+ hours wasted just figuring out task assignments
  • 🔴 Critical features forgotten because nobody tracked them
  • 🔴 Team members duplicate work due to poor communication
  • 🔴 Discord messages buried in endless scrolling
  • 🔴 "Who's doing what?" becomes the most asked question

I realized that Discord is where hackers live, but it's terrible for project management. Traditional project management tools feel like overkill for a 24-hour hackathon. We needed something that meets teams where they are - in their Discord servers - and automatically handles the boring stuff so they can focus on what matters: shipping code.

Current-AI was born from this frustration. It's the project management system I wish I had at my first hackathon.

What it does

Current-AI is an AI-powered project management system that lives inside your Discord server. Just chat naturally, and watch the magic happen:

🤖 Natural Language Task Creation

  • Type "I need an API for user authentication" in Discord
  • AI instantly creates a task, assigns it to backend devs, sets priority, and adds it to the Kanban board
  • No forms. No clicking. Just talk.

🎯 Smart Auto-Assignment

  • AI analyzes task content (not who requested it!)
  • Automatically assigns to the right team member based on their role
  • Example: "Design needed for landing page" → auto-assigned to designers

📊 Live Kanban Dashboard

  • Beautiful drag-and-drop board with 4 columns: Backlog, Todo, In Progress, Done
  • Real-time updates across all team members
  • Filter by team member to see individual workloads
  • Click any task to see full details, comments, and history

👥 Team Approval Workflows

  • Configurable approval requirements (features need 2+ votes, bugs auto-approved)
  • Proposal threads created automatically in Discord
  • AI-powered vote detection: just say "approved" or "LGTM"
  • Democratic decision-making for important features

✅ GitHub Auto-Completion

  • Commit message: "Fixed user authentication bug"
  • Current-AI: "Hey, that matches the 'Fix user auth' task! Marking as done."
  • Smart matching algorithm with confidence scoring
  • No manual status updates needed

🗣️ RAG-Powered Q&A Bot

  • Mention the bot: "@Current-AI who's working on the database?"
  • Get instant answers about your project
  • Context-aware responses using all project data
  • Like having a project manager who never sleeps

📝 Discord Intent Detection

  • "I'm working on the landing page" → Moves task to In Progress
  • "Finished the API integration" → Marks task as Done
  • "Delete the old task" → Removes task
  • "Put X in backlog for later" → Moves to Backlog
  • AI understands context and takes action

🔔 Real-Time Discord Notifications

  • Task status changes? Discord notification.
  • Someone completed a task? Team gets notified.
  • GitHub commit matched? Automatic update.
  • Keep everyone in sync, always.

📈 Activity Logging

  • Every action tracked with timestamp and user
  • See who did what, when
  • Audit trail for the entire project
  • Perfect for post-hackathon retrospectives

👤 16+ Role Types

  • Backend, Frontend, Full Stack, Designer, QA, DevOps
  • Data Scientist, Mobile, Security, Database Admin
  • Product Manager, Scrum Master, Technical Writer, and more
  • Precise task assignment based on specialization

💬 Task Comments & Updates

  • Add comments directly on tasks
  • Edit task details (title, priority, assignee, difficulty)
  • Full conversation history preserved
  • Collaborative task refinement

🎨 Smart Priority & Difficulty Detection

  • AI determines if task is Low/Medium/High priority
  • Assigns difficulty level: Easy, Medium, Hard
  • Helps teams tackle the right tasks first
  • No manual categorization needed

How we built it

Current-AI is a full-stack monorepo combining cutting-edge technologies:

🎨 Frontend

  • Next.js 16 with App Router for blazing-fast server/client rendering
  • React 19 with modern hooks and concurrent features
  • TypeScript for type-safe development
  • Tailwind CSS 4 for beautiful, responsive UI
  • @hello-pangea/dnd for smooth drag-and-drop Kanban
  • Lucide React for crisp, modern icons

⚙️ Backend

  • Next.js API Routes for serverless endpoints
  • MongoDB with native driver for flexible data storage
  • Session-based authentication with secure cookie management
  • bcrypt password hashing (crypto SHA-256)
  • RESTful API design with proper error handling

🤖 Discord Bot (Python)

  • discord.py 2.3.2 for Discord integration
  • aiohttp for async HTTP requests to Next.js API
  • python-dotenv for environment configuration
  • Real-time message listener with intent detection
  • Thread management for proposal voting

🧠 AI Engine

  • Google Gemini 2.5 Flash for natural language understanding
  • Intent classification: CREATE_TASK, UPDATE_STATUS, COMPLETE_TASK, DELETE_TASK, etc.
  • Smart role assignment using keyword analysis (150+ keywords mapped to 16 roles)
  • RAG (Retrieval-Augmented Generation) for context-aware Q&A
  • Fallback pattern matching when AI quota exceeded (no downtime!)

🔗 GitHub Integration

  • Webhook endpoint for push events
  • Commit message parser with keyword extraction
  • Smart matching algorithm (scores 0-1 confidence)
  • Auto-completion when commit matches task
  • HMAC signature verification for security

🗳️ Approval System

  • Keyword-based vote detection (no AI quota usage)
  • Configurable approval thresholds per task type
  • Auto-approval for task requester (you already want it!)
  • Team size validation (can't require more approvals than team members)

📦 Key Libraries

{
  "@google/generative-ai": "^0.24.1",
  "mongodb": "^7.0.0",
  "discord.py": "2.3.2",
  "clsx": "^2.1.1",
  "date-fns": "^4.1.0"
}

🏗️ Architecture Highlights

  • Unified user system: Auth users + manual team additions
  • Real-time polling: 3-second intervals for dashboard updates
  • Smart caching: Discord user ID → username mapping
  • Error resilience: Fallback patterns when services fail
  • Type safety: Full TypeScript coverage with strict mode

Challenges we ran into

🎙️ Voice Call Recording Feature (Abandoned)

The original vision included voice call recording and AI summarization to create tasks from verbal discussions. Imagine saying in a Discord voice call, "We need to implement OAuth" and having Current-AI automatically create that task.

What went wrong:

  • Discord.py voice integration required opus codec configuration nightmare
  • PyNaCl dependency hell across different operating systems
  • Audio chunking and buffering was way more complex than expected
  • Google Speech-to-Text API integration added another layer of complexity
  • Real-time transcription while processing was computationally expensive

After 8+ hours fighting dependencies and realizing I had <12 hours left, I made the tough call to cut the feature. Sometimes knowing when to pivot is the most important skill at a hackathon.

⚙️ Configuration Nightmares

  • MongoDB connection pooling in Next.js 16 App Router behaved differently than Pages Router
  • Discord.py intents - forgetting message_content=True made the bot completely deaf
  • CORS issues between Next.js API and Python bot (solved with proper headers)
  • TypeScript strict mode caught hundreds of potential bugs (but slowed initial development)
  • Gemini API quota limits - had to implement fallback pattern matching

⏰ Time Pressure

  • Built the entire system in under 48 hours
  • Had to prioritize: Core features over nice-to-haves
  • Spent 4 hours debugging why tasks weren't appearing (forgot to copy to /outputs directory in bot)
  • Last-minute styling fixes for the landing page

🐛 Unexpected Bugs

  • User filter infinite loop - onTaskCountsUpdate callback triggered re-renders
  • GitHub auto-completion matching too many tasks (had to add confidence threshold)
  • Proposal auto-approval showing 0/1 instead of 1/1 when only 1 approval needed
  • Input placeholder visibility - users couldn't see what they were typing (CSS color issue)

The lesson? Have a backup plan. When voice features failed, I focused on making text-based AI absolutely bulletproof.

Accomplishments that we're proud of

✅ We Actually Finished!

This isn't vaporware. Current-AI is fully functional, deployed, and working right now. You can sign up, connect your Discord, and start managing projects immediately.

🚀 It's Production-Ready

  • Zero critical bugs in final build
  • Stable performance with real-time updates
  • Graceful error handling throughout
  • Mobile-responsive design
  • Secure authentication with session management

🤖 AI That Actually Understands Context

Most "AI" projects at hackathons just slap GPT on top of something. We built:

  • Smart intent classification with 8 different intents
  • Role assignment algorithm using 150+ keywords
  • RAG system that actually retrieves relevant context
  • Fallback logic when AI services fail
  • Confidence scoring to prevent false matches

🎯 Solves a Real Problem

This isn't a solution looking for a problem. After 22 hackathons, I know this pain point intimately. Current-AI solves it elegantly.

🎨 Beautiful UX

  • YC-style landing page with gradient animations
  • Smooth drag-and-drop Kanban
  • Glass-morphism design elements
  • Thoughtful micro-interactions
  • Professional polish (not a typical hackathon MVP)

🧠 Smart Engineering Decisions

  • Unified user system: Seamlessly combines auth users + manual additions
  • Smart level badges: Only show when different from priority (reduces visual noise)
  • GitHub matcher: Sophisticated algorithm with keyword extraction and confidence scoring

🏆 Hope to Help Other Hackers

The greatest accomplishment would be seeing other hackathon teams use Current-AI and actually win because they spent time building, not organizing.

What we learned

🎓 First Discord Bot Ever

I had zero experience with Discord.py before this hackathon. Now I understand:

  • Bot intents and permissions
  • Message event handling
  • Thread creation and management
  • Webhook integration
  • Async Python patterns

Going from "What's a Discord bot?" to building one that does natural language processing in 48 hours was wild.

📊 First Project Management Project

Never built a Kanban board before. Learned:

  • Drag-and-drop libraries (@hello-pangea/dnd)
  • State management for real-time updates
  • Task lifecycle management
  • Activity logging patterns
  • User role systems

🤖 AI Integration at Scale

Using AI isn't just about calling an API. I learned:

  • Prompt engineering - getting consistent JSON responses
  • Fallback strategies - what happens when AI fails?
  • Cost optimization - when to use AI vs. simple patterns
  • RAG architecture - retrieving relevant context efficiently
  • Confidence thresholds - not all predictions are equal

🐛 Debugging Complex Systems

When 3 systems talk to each other, bugs get weird:

  • Added extensive logging (console.log is your friend)
  • Traced data flow across system boundaries
  • Used TypeScript to catch bugs before runtime
  • Learned to read Python stack traces
  • MongoDB queries debugging

💡 Most Important Lesson

Build for users, not for judges. I could've added flashy features that don't work. Instead, I focused on making something actually useful that solves a real problem.

What's next for Current-AI

Current-AI works great for hackathons, but it's currently hackathon-built. Here's the roadmap to make it production-grade:

🔐 OAuth Integration

Current state: Manual Discord IDs and GitHub webhook URLs
Future: One-click OAuth setup

  • Discord OAuth: Sign up with Discord, auto-detect servers, instant bot invite
  • GitHub OAuth: Connect repositories with 2 clicks, automatic webhook setup
  • Google OAuth: Optional integration for team email notifications

🌐 Multi-Server Support

Current state: One project per deployment
Future: Multiple teams, multiple projects

  • SaaS model with tenant isolation
  • Per-team billing and analytics
  • Admin dashboards per organization
  • Team invite system via email/Discord

🎙️ Voice Call Integration (Revisited)

Now that the core is stable, time to tackle this properly:

  • Discord voice channel listener
  • Real-time transcription
  • AI summarization of meetings
  • Automatic action item extraction
  • "Meeting notes" feature

📊 Advanced Analytics

Current state: Basic activity logs
Future: Data-driven insights

  • Velocity tracking (tasks completed per day)
  • Burndown charts
  • Team contribution metrics
  • Bottleneck detection
  • Predictive completion estimates

🤝 Third-Party Integrations

  • Figma: Link designs to tasks
  • Vercel/Netlify: Deploy status → task completion
  • Linear/Jira: Two-way sync (for teams transitioning)
  • Google Calendar: Deadline reminders
  • Notion: Documentation links

🎨 Customization

  • Custom Kanban columns (not just 4)
  • Custom task types (beyond bug/feature/todo)
  • Team-specific role types
  • Custom AI prompts per team
  • Theme customization

Built With

Share this project:

Updates