Enterprise AI Agent Platform
Blue Star Group's internal AI portal: one shared platform that solves once what every agent needs —reaching the data, searching, measuring and governing— instead of rebuilding it in each project.
The problem
Daily business decisions —pricing, purchasing, which products to sell— relied on manual analysis repeated in each of the company's 11 countries, with no shared data platform and no standard way to measure impact.
What I did
Founded the AI Lab from scratch: hired and led a team of up to 3, defined the Lab's technical standards (reference architecture, Docker deploy patterns) and led the ecosystem's shared platform: an MCP data-access pilot, a shared vector layer, adoption telemetry with its dashboard and an AI governance framework applied org-wide.
The outcome
20+ LLM agents and decision engines in production on one platform, for Todomoda and Isadora's operation over data from 11 countries. Also defined the Lab's secrets-handling and secure-deploy standard.
How it's built
- Data from 11 countriesInput
- Data access over MCPOne path to the data
- Shared vector layerSearch by similarity, not by code
- Adoption telemetryHow impact is measured
- AI governance frameworkApplied org-wide
- 20+ agents in productionOutput
One shared platform instead of one agent per problem: data access, the vector layer, telemetry and governance are solved once, and every agent builds on them.
Stack
- Python
- Streamlit
- Docker
- MCP
- Qdrant
- Supabase
- SQL Server
- OpenAI
- Gemini
- Claude
Blue Star Group internal project — the code is not public.
Next project: Intelligent Markdown Engine →