aswin-shine.dev · portfolio and project case studies

DevOps/MLOps engineer focused on AWS EKS, Terraform, and GitOps delivery, with a design-and-frontend background that shapes how I build developer- and user-facing tooling. I've taken infrastructure from bare cloud to production: Kubernetes clusters via IaC, CI/CD pipelines with security gates, and observability stacks - alongside a RAG chatbot built on LangChain and Pinecone and deployed to AWS EC2. Comfortable owning the stack end to end: Linux servers, Nginx/SSL, Docker, and automated pipelines.

Currently deepening my System Design and scalable-architecture skills.

Penwave - Cloud-Native Blogging Platform on AWS (EC2 and EKS) (repo · case study) Full-stack blogging platform (Next.js, Node.js/Express, PostgreSQL, Redis) that ran in production on EC2 with Docker Compose, then was rebuilt as a modular monolith on AWS EKS with cluster and networking provisioned via Terraform (tested, verified, and spun down for cost).

• Packaged as multi-stage, multi-arch (amd64/arm64) Docker builds deployed via Helm across frontend, backend, and ingress in a GitOps workflow with ArgoCD, with HPA (2-6 replicas) and Prometheus/Grafana/Loki for monitoring and centralized logging.
• Hardened with Argon2id password hashing, JWT refresh-token rotation, Redis-backed rate limiting, Zod-validated inputs, and Trivy image scanning in CI/CD, fronted by Nginx with Let's Encrypt TLS termination.

Formula 1 RAG Chatbot - AI Q&A System (repo · case study) Conversational RAG system over historical F1 data (races, drivers, results, standings) and technical PDFs, with real-time web-search fallback.

• Hybrid retrieval: Pinecone vector search (k=3, cosine, 384-dim HuggingFace embeddings) with a context-sufficiency threshold that falls back to SerpAPI web search when local context is thin, orchestrated via LangChain's RunnableWithMessageHistory for multi-turn conversational memory.
• Shipped via a 3-stage GitHub Actions pipeline (integration → delivery → deployment) to Docker Hub, deployed on a self-hosted AWS EC2 runner serving GPT-4.1-mini responses through a Flask API (since retired).

YouTube Sentiment Insight - MLOps Pipeline & Chrome Extension (repo · case study) End-to-end MLOps system for YouTube comment sentiment classification, from DVC-orchestrated training to a production Chrome extension.

• Built a 5-stage DVC pipeline (ingestion → preprocessing → model building → evaluation → registration) training a LightGBM classifier on TF-IDF features, tracked via MLflow with S3 as the remote/artifact store, reaching 85.3% accuracy / 84.9% F1 on the test set.
• Automated a 3-stage GitHub Actions pipeline (CI → Docker build/push to Docker Hub → CD to AWS EC2 via self-hosted runner) serving predictions through a Flask REST API, consumed by a Chrome extension for real-time comment scraping, sentiment visualization, and trend analysis.

Cloud, Infrastructure & Containers

Backend, Databases & Frontend

Not yet on skillicons.dev - ArgoCD, Helm, Prometheus, Grafana, Loki, SonarQube, Trivy, LangChain, Pinecone, HuggingFace, MLflow, OCI:

• Oracle Cloud Infrastructure 2025 Certified Foundations Associate - Oracle
• Introduction to Containers, Kubernetes, and OpenShift (CC0201EN) - IBM / Cognitive Class
• MlOps Zero to Hero - Udemy
• Ethical Hacking from Scratch - Udemy
• Introduction to Industry 4.0 and Industrial Internet of Things - NPTEL / Swayam
• Introduction to Professional Scientific Communication - NPTEL / Swayam