Shreyansh Sancheti

AI Infrastructure Engineer at Microsoft Azure Compute. I build open-source infrastructure for AI agents, in public. This is my showroom.

Scroll to walk through. Click anything on display.

Experience, parked in order: the oldest on the left, the newest on the right. Each car is a chapter.

• Red Hat1957 Bel Air
• IBM1967 Impala
• Arista Networks2019 Mustang Roush
• XBattery1970 café motorcycle
• Microsoft2026 Rolls-Royce Spectre

1957 Bel Air · Where it started.

Associate Software Engineer, Distributed Systems

Jan 2022 – Dec 2022

• +25%storage throughput
• -30%production incidents
• 100+code reviews

• Improved distributed storage throughput 25% via automated profiling and bottleneck analysis.
• Cut production incidents 30% with monitoring and alerting; cut QA cycle time 40% with Docker-based performance test automation; 100+ code reviews.

1967 Impala · Built to go the distance.

Software Engineer II, Cloud Storage & Automation

Jan 2023 – Feb 2025

• -60%MTTR
• 1000+deployments watched
• 10xbenchmark productivity

• Built Ceph performance monitoring (Prometheus, Python) detecting anomalies across 1000+ enterprise deployments.
• Automated troubleshooting with Python/Bash log analysis, cutting incident response time 40%.
• Built Rook-Ceph Kubernetes operators, cutting MTTR 60%; 15+ PRs merged to Ceph.
• Built Docker/Ansible S3 benchmarking framework, a 10x productivity gain.

2019 Mustang Roush · Light, fast, precise.

Software Engineer, Network Infrastructure & Performance (C++)

Feb 2025 – Feb 2026

• 100%packet loss eliminated
• -40%MTTR
• +35%deploy reliability

• Built BESS agent anomaly detection, eliminating 100% packet loss during port updates.
• Automated incident detection and RCA correlation, cutting MTTR 40%.
• Built Grafana dashboards for 400G+ systems; CI/CD raised deployment reliability 35%.

1970 café motorcycle · The side road, ridden at the same time.

Senior Infrastructure Consultant, AI & Cloud Platform

Mar 2025 – Oct 2025

• 99.9%uptime
• 10,000+SMEs served
• -75%false positives

• Built AI Kubernetes observability platform for 10,000+ SMEs: 99.9% uptime, 60% less manual work.
• Delivered Terraform/Ansible/GitLab CI pipelines enabling 50+ daily zero-downtime deploys.
• Correlated alerts across 500+ VMs, cutting false positives 75%; automated RCA resolved incidents 50% faster.

2026 Rolls-Royce Spectre · The flagship.

Software Engineer II, Container Platform & Infrastructure (Azure Compute / Hyper-V)

Mar 2026 – Present

• HCSAPI migration
• LiveMigration at Azure scale

• Drive HCSShim/HCS API migration to modern container abstractions, enabling Live Migration at Azure scale.
• Built parity/regression validation frameworks across container runtime layers.
• Built PowerShell/Go deployment tooling and CI/CD for rollout and validation.
• Root-cause distributed failures across Azure clusters using Kusto telemetry.

Every tool I reach for, kept on the wall. The warm ones are the ones I use most: Kubernetes, vLLM, Go, Rust, Terraform.

Languages

Python, Go, Rust, C/C++, Swift, Bash, PowerShell, TypeScript, SQL

AI Infrastructure

LLM inference, vLLM, KServe, llm-d, Inference autoscaling (HPA), Model serving, Multi-tenant isolation, MCP, RAG, LangGraph, LangChain, Azure OpenAI, Agentic workflows

Cloud & Platform

Kubernetes, Operators & CRDs, Azure, AWS, Docker, Terraform, Ansible, containerd, Hyper-V, gVisor, Cilium, Ceph, Linux

Observability & Delivery

Prometheus, Grafana, OpenTelemetry, Kusto, ELK, GitHub Actions, GitLab CI, Incident response, RCA, High availability, Cost optimization

Customer-facing

Solution design, Enterprise integration, Technical advisory, REST/gRPC APIs, n8n workflows, Human-in-the-loop AI, Demos, Conference talks

Projects I build in public. Flip through them here or on the lightbox.

The get-to-know-me room. A good suit, a few watches, and what I do when the laptop is shut.

• 

  Badminton

  Fast hands, faster feet.

• 

  Cricket

  Weekend overs, never enough of them.

• 

  Guitar

  The one thing I play slowly on purpose.

Kept for the craft. Every one of them is set to Bengaluru time. Pick one up to turn it over.

• Patek Philippe→
• Rolex Datejust→
• Omega Seamaster 300→
• Seiko→

• SREcon26 EMEA

  Speaker · Dublin, Oct 2026

  Why the Kubernetes autoscaler fails LLM inference

• Ceph Days

  Speaker

  Distributed storage, from the people who maintain it

• Agentic Summit

  Speaker · Bangalore

  AI agents on real infrastructure

• 15+ merged PRs

  Ceph upstream

  RADOS Gateway and storage engine work

• Mercatus grant

  Agentic Operator

  Kubernetes governance for AI agents

• 52 stars

  agentic-operator-core

  Open source, built in public

A parting thought