Arjun Ganesh

Python, CUDA C++, TypeScript. Azure AI Foundry + Semantic Kernel, AWS Bedrock/Lambda, Google ADK + Vertex AI. A2A and MCP, FastAPI, Postgres/pgvector, CockroachDB, Kubernetes, CuPy and llama.cpp.

CUDA-Q and cuQuantum, after benchmarking simulated amplitude estimation against a hand-written CUDA kernel. Also Google's ADK and A2A in anger — deployed across services, not run locally.

Bastion — governed ADK agents auditing the live IAM of the project running them, read-only by design. Alongside q1729 and llm-qlab: measured CUDA, quantum-simulation and GGUF quantization benchmarks.

Agent architecture across Semantic Kernel, ADK, A2A and MCP; least-privilege identity for autonomous systems; GPU benchmark methodology. Also: why your agent works locally and dies in production.