SupportMind
SupportMind is an AI-powered support assistant with long-term memory built using CockroachDB and AWS.
Inspiration
Support engineers constantly lose important context between conversations. We wanted to build an assistant that remembers previous interactions, retrieves relevant knowledge when needed, and can inspect its own database through CockroachDB MCP tools.
What it does
SupportMind combines three types of memory:
- Recent conversation context
- Long-term semantic memory stored in CockroachDB
- Live database inspection using the CockroachDB Managed MCP Server
The assistant automatically stores important user information, retrieves relevant memories using vector search, and can answer database-related questions by executing MCP tools.
How we built it
The project consists of:
- React frontend
- FastAPI backend
- OpenAI API
- CockroachDB Cloud
- CockroachDB Distributed Vector Search
- CockroachDB Managed MCP Server
- AWS Amplify deployment
Conversation memories are embedded and stored inside CockroachDB. Before every response the assistant retrieves recent context, performs semantic search, optionally executes MCP tools, builds a complete prompt, generates the answer, and stores new memories.
Challenges
The biggest challenge was combining semantic retrieval with deterministic MCP tool execution while keeping the workflow simple and transparent.
Another challenge was designing a memory pipeline that separates recent conversation history, long-term memory retrieval, and live database inspection.
What we learned
During this project we learned how to combine vector databases, agentic workflows, MCP servers, and cloud deployment into a single AI application.
What's next
Future work includes integrations with Help Desk systems, Slack, Telegram, CRM platforms, and multi-tenant organization support.
Built With
- alembic
- amazon-web-services
- amplify
- cockroachdb
- docker
- fastapi
- github
- mcp
- openai
- pydantic
- python
- react
- restapi
- sqlalchemy
- typescript
- vite
Log in or sign up for Devpost to join the conversation.