Software Engineer · Systems & Infrastructure
Distributed systems, Kubernetes tooling, and the kind of infrastructure work that runs underneath production traffic. Go on the backend, Terraform for everything that provisions, AWS underneath.
Most engineering bios pick a lane. Mine doesn't.
I build infrastructure tools. Chaos engineering, Kubernetes controllers, SLO monitors, Terraform-with-an-LLM-on-it. Go on the backend. Terraform underneath. AWS underneath that.
The argument below isn't that any one of these is groundbreaking. It's that the same person built all of them.
B.S. Computer Science, Adelphi University, 2025. Based in New York. SRE, DevOps, and Platform Engineering. Work authorized on OPT.
Infrastructure should be boring. Boring means nothing's on fire.
— working principle
$ kubectl get pods
payment-api-7f2 0/1 CrashLoopBackOff
+ resource "aws_s3_bucket" "logs" { + versioning = true + encryption = "aws:kms" + }
sum(rate(http_requests_total{status="5xx"}[5m]))
→ 0.034 req/sec · within SLO
replicas: 3
resources:
limits:
memory: 512Mi
requests:
memory: 128Mi
The idea behind chaos engineering is straightforward: break things deliberately so they stop breaking by surprise. Havoc runs on AWS EKS and lets you schedule experiments, pod kills, network latency injection, and CPU pressure through a CLI that dispatches commands via Kafka to agents on every cluster node. Each agent enforces a blast radius limit so nothing gets out of hand, and a Redis kill switch stops everything cold if it does. Every result lands in a Postgres ledger. The whole stack is provisioned with Terraform and deploys via Helm.
View live →You describe the infrastructure you want in plain English. TerraSense generates the Terraform HCL. What makes it more than a code generator is the correction loop: it runs Checkov on every plan, feeds the policy violations back to the model, and keeps rewriting until the plan is clean. Nothing runs terraform apply without a human approving it first. Every draft, violation, and correction is logged to Postgres, so there's always a full audit trail.
View on GitHub →I wanted to compare cricket players across T20, ODI, and Test formats using real match data. The backend is FastAPI over PostgreSQL with 2,962 international matches loaded in. Midway through the build, my API provider hit its free-tier limit, so I switched ingestion to Cricsheet and rewrote just that layer. The schema, API, and frontend stayed untouched. It's live.
View live demo →When a pod enters CrashLoopBackOff, someone usually gets paged at 3 AM to run kubectl describe and figure out what happened. Kube-Sentinel handles that loop automatically. It watches every pod via a streaming watch, reads exit codes and logs to classify the failure, and applies the right fix — memory patch for OOM kills, restart-then-rollback for repeated crashes, alert-only for config errors (those still need a human). Runs as a non-root distroless container with Prometheus metrics on :8080.
View on GitHub →A service that hits your HTTP endpoints on a schedule and asks: Are we meeting our SLOs? It tracks availability and latency compliance over a rolling window and calculates how much error budget you have left. When a threshold is crossed — budget below 50%, five failures in a row, budget gone — it fires one alert. Not one per minute. One. The alert is structured JSON, so you can pipe it to PagerDuty or Slack with a single command. Ships as a ~15MB Docker image.
View on GitHub →A networked chess game where two players connect to a server, get paired, and play in real time. The server handles up to 20 concurrent matches simultaneously, each isolated in its own thread. The game engine validates every legal move using FEN-based board state — castling, en passant, and pawn promotion included. Client is built in JavaFX, communication runs over TCP sockets with object serialization.
View on GitHub →Amazon Web Services
HashiCorp
Oracle
Linux Foundation
Type a command. Or click one of the floating chips above. This is a static mock — no real cluster runs underneath.