Senior Member of Technical Staff @ Oracle — OCI Functions
12+ years building hyperscale cloud infrastructure · prev. Microsoft Azure & Bloomberg · IEEE Senior Member
I build and operate hyperscale cloud platforms. Currently I define technical strategy and architecture for OCI Functions — Oracle's Kubernetes-based serverless platform — including its multi-architecture (x86 + ARM) runtime migration. Daily tools: Go, Python, Java, Kubernetes, Terraform, and gRPC.
Previously: Microsoft Azure (real-time Kafka ingestion pipelines; security infrastructure for Azure Networking), Bloomberg (0→1 cloud-native alerting platform — Go, Python, AWS Lambda, PostgreSQL, OpenTelemetry, Kubernetes), and OCI Telemetry (Kafka-based routing for Oracle's observability data plane).
On the side, I build open-source AI-infrastructure tooling — model routing, agent observability, cluster guardrails, and reproducible ML data pipelines — each shipped with a citable preprint (below).
AI infrastructure · LLM inference & agents · Kubernetes · serverless · platform engineering · distributed systems · Kafka / streaming · observability
Each project pairs working code with a citable preprint — design decisions and claims written down where they can be checked.
| Project | What it is | Paper |
|---|---|---|
| switchboard | Privacy-aware, local-first router for CLI coding agents (Codex, Claude Code) and local LLM inference — keeps sensitive prompts on-device. Benchmarked in the preprint: 62% fewer premium-agent calls at 4.1/5 vs 4.6/5 always-premium quality, with zero privacy leaks | |
| agent-tracebench | Reproducible observability, replay, and regression checks for LLM agents | |
| kube-clusterguard | Static guardrails for Kubernetes AI/ML compute clusters | |
| shardflow-ml | Deterministic manifest, planning, and checkpoint layer for reproducible ML data pipelines |
Also: ai-spend-cap-tracker — a sourced, public tracker of organizations capping or cutting employee AI-coding spend; the cost pressure switchboard is built for.
- OpenTelemetry — merged:
opentelemetry-instrumentation-pikaduplicate consumer-span fix - HeadRoom — LLM context-compression proxy; fixes across its proxy, router, and compression paths → my merged PRs
- gRPConf North America 2026 (The Linux Foundation / CNCF) — “gRPC Lessons From Serverless Infrastructure” · Mountain View, CA · Sep 3, 2026 · session
- Asynchronous Data Stream Ingestion in Distributed Cloud Infrastructure — IJCTT, 2024 · DOI
- IEEE Senior Member · Panelist, IEEE Senior Member Application Review Panel
- Peer Reviewer, IEEE Transactions on Cloud Computing · ACM Computing Surveys · Journal of Open Source Software
- Professional Member, British Computer Society (MBCS)
📫 Reach me: ai.vinaygupta@gmail.com · linkedin.com/in/aivinay