May 2025 – now
AI Engineer · Sekera Group, Casablanca
A multi-tenant security-operations platform. I work on its agent layer, from the first version to running it in production.
- Production agents on the Claude Agent SDK: custom MCP tools, subagents, sessions you can steer mid-turn, human approval of actions, background jobs under budget limits.
- Observability: token, cache and cost accounting per session, a live view of turns in flight, job tracing.
- Multi-provider LLM gateway: five backends behind one interface, normalized streaming, encrypted bring-your-own-key credentials.
- Sandboxing and back end: every agent session runs in a gVisor sandbox with no secrets inside; async FastAPI services on PostgreSQL, Docker, CI/CD.