Enterprise AI · Enterprise Productivity
NovaFloat: Enterprise AI Productivity Platform
AI platform unifying 1500+ prebuilt connectors, UI-attached custom MCP servers and a visual workflow builder, delivering a 65%+ productivity lift across departments.
- Role
- AI Solution Architect
- Architected and built the platform end to end, from connector layer to agent harness.
- Industry
- Enterprise Productivity
- Timeline
- Delivered - in production
Business impact
- productivity lift
- 65%+
- prebuilt connectors
- 1500+
- teams onboarded
- HR, DevOps, Infra, cross-functional
- custom tooling
- Python MCP servers from UI
Overview
AI platform unifying 1500+ prebuilt connectors, custom MCP servers authored in Python and attached straight from the UI, and an n8n-style visual workflow builder for bespoke automations.
On top of any connector, users compose knowledge bases and AI pipelines, while a harness-engineered agent reasons across the relevant connectors, MCP tools, knowledge bases and live web search, the platform's core interactive layer. Backend engineered in FastAPI with Microsoft Entra ID OAuth, PostgreSQL for application state, Azure AI Search for knowledge bases and Redis caching. Rolled out to HR, DevOps, infrastructure and cross-functional teams, lifting productivity 65%+ across departments.
Technical architecture
- Connector Layer
- 1500+ prebuilt connectors plus UI-attached custom MCP servers
- Agent Harness
- Harness-engineered agent reasoning across tools, knowledge and web search
- Workflow Builder
- Visual, n8n-style builder for bespoke automations
- Backend
- FastAPI with Entra ID OAuth, PostgreSQL state and Redis caching
- Knowledge
- Azure AI Search-backed knowledge bases per connector
Key features
- 1500+ prebuilt connectors available out of the box
- Custom MCP servers authored in Python and attached directly from the UI
- n8n-style visual workflow builder for bespoke automations
- Knowledge bases and AI pipelines composable on top of any connector
- Harness-engineered agent reasoning across connectors, MCP tools, knowledge bases and live web search
- Microsoft Entra ID OAuth with per-team scoping
Challenges solved
- Making custom MCP servers safe to author and attach from the UI without a deployment step
- Keeping a single agent coherent while reasoning across 1500+ possible tool surfaces
- Scoping knowledge bases and connectors per team without duplicating infrastructure
Tech stack
- Python
- FastAPI
- Model Context Protocol (MCP)
- Harness Engineering
- Microsoft Entra ID OAuth
- PostgreSQL
- Azure AI Search
- Redis
- Azure Container Apps
- Workflow Orchestration
- Azure OpenAI
- Docker