WorkSpicychat (Nextday)Draft, not yet approved
Memory, moderation and faster releases for Spicychat
We led the backend team at Spicychat, where users create and publish their own AI chatbots. We improved semantic memory, built an AI-plus-human moderation workflow that made moderation 80% more efficient, and sped up releases by 30%.
- More efficient bot moderation cycle
- 80%
- Faster release cycle for features and deployments
- 30%
- Off backend API response times
- ~5ms
Challenge
On Spicychat, users create AI chatbots with their own personalities and publish them for other people to talk to. Nextday, the company behind it, needed three things fixed at once. Conversations had to get better. The community had to stay safe on a platform where anyone can publish a bot. And the release cycle had to speed up.
What we built
We led a team of three on the backend.
For conversation quality, we improved the bots' semantic memory so dialogue got richer. For safety, we built a bot moderation workflow that pairs AI moderation with human review. For the team, we reworked the release cycle for features and deployments, which also made the developers more productive.
How
The backend is Node.js and TypeScript microservices on AWS, with DynamoDB for storage. We also spent time on API design and endpoint latency, which sped up bot deployments and updates.
Results
- Bot moderation became 80% more efficient.
- Features and deployments ship 30% faster.
- Backend API response times dropped by about 5ms.