Jan 2026 - Apr 2026
Cloud2 Labs
Open Source Developer
Built and maintained open-source AI blueprints across multiple inference environments, with work spanning benchmarking, multi-agent orchestration, authorization, and repository quality automation.
Role Overview
A quick read on what the role was and the main areas I worked on before the deeper implementation details.
Owned the open-source Innovation Hub repository with 15 AI blueprints, independently built 4 of them, and handled updates, fixes, and feature work across the broader platform.
Built and maintained AI workflows across OpenAI, VLLM, Ollama, and Intel OPEA so the same blueprints could be benchmarked across different model and serving environments.
Designed multi-agent systems with CrewAI and LangGraph, including planner-led routing, intent classification, low-confidence clarification handling, and selective agent communication.
Built security-focused authorization workflows with MCP, RBAC, and ABAC so protected data access was resolved deterministically before any LLM call.
Maintained repository quality and security automation with GitHub Actions, Trivy, and Dependabot to keep the open-source platform stable as it expanded.
Detailed Work
The role is grouped by major workstreams, with each one kept concise enough to explain the implementation and impact without turning into a wall of text.
Innovation Hub Ownership
Owned the open-source Innovation Hub to build and maintain AI blueprints that could run across different inference environments without turning each one into a separate implementation.
Managed the central repository for 15 AI blueprints, independently built 4 of them, and owned ongoing updates, fixes, and new feature work across the broader set.
Designed blueprint workflows to run across OpenAI, VLLM, Ollama, and Intel OPEA so the same applications could be tested against different model capabilities and serving constraints.
This made the Innovation Hub a practical benchmarking and showcase platform instead of a collection of one-off demos tied to a single inference setup.
Multi-Agent Orchestration
Designed a multi-agent blueprint for a transportation use case to show how planner-based routing, controlled agent communication, and cross-inference reliability could work in a practical workflow.
Used a transportation use case to design a broader multi-agent blueprint with CrewAI, where a planner agent routed requests across task-specific agents instead of pushing every request through the same reasoning path.
Implemented an intent taxonomy, low-confidence clarification handling, and selective agent-to-agent communication so the workflow stayed controlled, avoided unnecessary LLM calls, and reduced token and context waste.
This made the blueprint easier to benchmark across different inference environments because the orchestration logic was structured, portable, and not dependent on one model or serving setup to work well.
Authorization and Access Control
Built a security-focused blueprint to show how authorization could be enforced deterministically before any LLM call, instead of relying on a model to infer access rights.
Designed the blueprint around RBAC and ABAC rules so document access could be resolved correctly across different departments, roles, and access levels.
Built an authorization MCP service that returned structured access decisions before the planner, summarization, or analytical agents could access protected data.
This created a safer and more reusable pattern for agent workflows because access control stayed deterministic, auditable, and independent from model behavior.