
Initiatives / Systems & Infrastructure
Tamarin
One operational view of Discord automation and AI activity for Monash Automation: workflows, inductions, points and an internal AI agent called Hermes.
- Team
- Systems & Infrastructure
- Project lead
- Midhun Kumar
- Stack
- Next.js, Django, py-cord, ChromaDB, llama.cpp
Structure and solidity
Tamarin brings Discord workflow automation, AI agent execution, lab inductions and the points leaderboard into a single dashboard. A workflow moves from trigger definition through its chain of actions into live execution, moderation and Notion sync without losing its trigger type, action order or owning admin. It removes the need for one-off Discord bot code per request and keeps a clear record of how each workflow run, induction or point award happened.
Foundation and development
The dashboard is built with Next.js, React and TypeScript, authenticated through NextAuth with Discord OAuth and optional Authentik SSO linking. The backend is a Django application exposing a REST API and running a py-cord Discord bot in a background thread of the same process. Notion is the source of truth: workflow definitions, induction records and points in Django are treated as a cache, and approvals in the dashboard sync back to Notion bidirectionally.
Agentic and intuitive
Hermes is Tamarin's resident agent. It answers @mentions in Discord, writes and ships code changes, keeps long-term memory across conversations, and drafts and shares documents on request. Workflows can also spin up scoped, tool-using AI agents that call a restricted set of workflow actions and bridge out to external MCP tool servers. Responses are grounded by a Notion-backed RAG pipeline that embeds organisational knowledge into a ChromaDB vector store, served through Tamarin's own OpenAI-compatible AI router in front of the team's self-hosted llama.cpp models.

