Current-AI

🏆 Best Use of MongoDB Atlas at Hacks for Hackers 2026. Built Current-AI, an AI-powered project management platform that converts Discord conversations into structured tasks with automated role assignment, real-time Kanban boards, GitHub-based auto-completion, and RAG-powered Q&A, eliminating coordination chaos for hackathon teams.

Anyone who has done a hackathon knows the real enemy is not the code, it is the coordination. Priorities get argued in Discord, someone volunteers for a job nobody writes down, and an hour later half the team is building the same thing while the other half waits on a blocker no one flagged. Current-AI was built to kill that chaos. It is an AI-powered project manager that lives in your Discord, quietly listening to the conversation and turning it into an organized, tracked set of tasks in real time. It won Best Use of MongoDB Atlas at Hacks for Hackers 2026. ### From chat to task, automatically The core of Current-AI is natural language task detection running on Google Gemini 2.5 Flash. As people talk in Discord, it reads the intent behind each message and decides on its own whether a task should be created, updated, completed, or deleted, so nobody has to stop and file a ticket. Each new task is then routed to the right person, matched across more than 16 developer specializations, which means work lands with whoever is actually suited to it rather than whoever happened to speak up first. ### A board that stays in sync All of that flows onto a real-time Kanban board built with Next.js 16 and React 19. Tasks sync live for everyone at once, support drag-and-drop, and carry priority and difficulty levels along with smart badges that surface only what matters. For the bigger, higher-stakes calls, there is an approval workflow: a proposal gets discussed and voted on inside the Discord thread before it is ever allowed onto the board, so the team stays in agreement without leaving the chat. ### Closing the loop with GitHub Current-AI also connects planning to actual execution through GitHub Webhooks. A fuzzy-matching algorithm watches incoming commits and works out which existing task each one relates to, and once it is confident enough, it marks that task complete on its own, no manual check-off required. Every one of these actions is written to a full audit log, so the team always has a transparent record of what changed, when, and why. ### Just ask the project Instead of scrolling threads to piece together where things stand, you can simply ask. Current-AI includes a Retrieval-Augmented Generation assistant that pulls context from the tasks, logs, and assignments stored in MongoDB Atlas and answers questions conversationally, things like who is blocked, what is in progress, or what is still left to do, right from Discord or the web dashboard. ### Under the hood The backend runs on Next.js API routes with secure session-based auth, paired with a dedicated Python Discord bot that handles live message ingestion, thread monitoring, and linking Discord accounts to users. Efficient queries and Redis caching keep it responsive even during the message firehose of a live hackathon. In practice, Current-AI cut organizational overhead by more than 60%, letting teams spend their time building instead of babysitting a task list. Next up is broader enterprise support, deeper integrations with other collaboration tools, and predictive task planning that anticipates work before it is even asked for.

At a glance

Timeline: Jan 2026
Type: Hackathon Project
Event: Hacks for Hackers 2026
Award: Best Use of MongoDB Atlas

Technologies

Next.js 16
React 19
TypeScript
Tailwind CSS 4
MongoDB Atlas
Next.js API Routes
Python 3.11
Discord API (discord.py)
Google Gemini 2.5 Flash
Retrieval-Augmented Generation (RAG)
GitHub Webhooks
Redis (Caching)
Natural Language Processing (NLP)

Photos

Current-AI announced as the Best Use of MongoDB Atlas winner during the MLH award stream

Links

Machine-readable version of this page: https://www.sherwinvishesh.com/project/current-ai.md | site index: https://www.sherwinvishesh.com/llms.txt