Shashank Sinha

Build a primary platform across all our groups and other suppliers. I helped with the following:

• Built and improved CI/CD tooling and helped developers with CI configurations.
• Managed CI for automated testing of all environments.
• Managed CD for automated cross-region deployment to multiple environments

Role: DevOps engineer

Designed and deployed a full-stack large language model (LLM) observability platform on Azure Kubernetes Service (AKS) for an AI-powered code analysis product. The platform provides end-to-end visibility into LLM behavior, latency, token usage, and error rates across all AI inference pipelines.

Integrated Langfuse (open-source LLM observability) with the existing OTEL Collector, Grafana, Tempo, Loki, and Prometheus stack to create a unified observability layer spanning both infrastructure and AI model layers. Resolved a critical multi-tenant Loki misconfiguration that had silently dropped logs for multiple teams for over six weeks. Reduced Sentry alert noise by 82% through intelligent grouping, enabling engineers to focus on genuine incidents. Built custom Grafana dashboards for LLM latency percentiles, token cost tracking, and trace correlation between model requests and infrastructure events.

The platform now serves as the observability foundation for all AI workloads, enabling data-driven prompt optimization and SLA enforcement for LLM endpoints.

STACK
Langfuse, OpenTelemetry, Grafana, Tempo, Loki, Prometheus, Sentry, Azure Kubernetes Service (AKS), Kubernetes, Helm.

Acted as the senior DevOps engineer for an eCommerce platform personalized for grocery retailers. Helped developers by providing them with ephemeral testing environments, implementing and improving platform stability and availability.

Built an AI-powered DevSecOps automation platform integrating Claude AI's Model Context Protocol (MCP) with enterprise security tooling to eliminate manual toil across the software delivery lifecycle.

CORE COMPONENT
• An AI agent that reads SonarQube blocker issues, generates targeted code fixes using large language models (LLMs), and automatically opens pull requests, cutting code quality remediation effort by over 80%. The agent understands code context, applies language-appropriate fixes, and routes findings through a structured review workflow.

SECOND COMPONENT
• Automated attack surface triage pipeline (Chariot scanner to Tines to GitHub ownership lookup to Jira and Bugcrowd routing) that processed 400+ security findings across 24 engineering teams, replacing weeks of manual triage with fully automated routing and ticket creation.

Built as a reusable Claude skill framework deployable across any engineering org. Represents the next evolution of DevSecOps: AI agents handling routine security toil autonomously.

STACK
Claude MCP, Python, SonarQube API, GitHub API, Jira API, Tines, Bugcrowd, Chariot.

I led the design and implementation of a distributed ordering platform infrastructure for a restaurant-tech startup, which processes millions of transactions per month. The project involved architecting a scalable, cloud-agnostic infrastructure across AWS, Azure, and Google Cloud Platform (GCP) using Terraform and Ansible, provisioning Kubernetes clusters, managed databases, and messaging layers. I built a high-fidelity staging environment that mirrored production, allowing for testing of integrations with over 40 POS systems and 70+ 3rd-party ordering services. I implemented end-to-end monitoring and alerting to achieve 99.99% uptime and optimized cloud resources to reduce monthly spend by 20%. This work improved release confidence, accelerated deployment cycles, and enabled remote teams across four continents to collaborate effectively.