type SamarthSharma struct {
    Education         map[string]string
    CurrentFocus      string
    Philosophy        string
    PrimaryLanguages  map[string]string
}

func main() {
    me := SamarthSharma{
        Education: map[string]string{
            "B.Tech": "IT @ GGSIPU",
            "B.Sc":   "Data Science @ IIT Madras",
        },
        CurrentFocus: "DevOps automation + ML performance",
        Philosophy:   "Understand first. Automate later.",
        PrimaryLanguages: map[string]string{
            "Go":     "services that need to be fast and correct",
            "Python": "ML work and anything I need to iterate on quickly",
        },
    }
}

I'm Samarth Sharma. I build backend systems and infrastructure that I actually understand.

I don't use tools I can't debug, and I don't trust abstractions I can't explain.

Right now, I'm bridging the gap between DevOps automation and ML performance optimization, from Terraform scripts to CUDA kernels.

mlops-tryops is my current main project — a production MLOps platform on EKS. FastAPI inference service with hot-reload from S3, ONNX runtime for serving, a training CronJob, Kyverno admission control with Cosign image signing, Prometheus + Grafana monitoring, drift detection, and Terraform for the entire stack. OIDC everywhere, no static credentials. Most of my recent learning has happened here.

Also grinding CUDA via ML-style problems. GPU memory management still trips me up.

I care about systems that don't break at 3 AM and code I can debug without Stack Overflow.

What gets me excited is when backend logic, infrastructure, and ML converge into something that actually solves a real problem, not just a demo.

MLOps Platform Production MLOps platform on EKS. FastAPI + ONNX inference with S3 hot-reload, daily training CronJob with auto-promotion, Kyverno admission control with Cosign image signing, and Terraform managing everything from VPC to GitHub secrets.	SlotSwapper Calendar swap platform with a Go backend and Vue 3 frontend. The core problem was making swap operations atomic and race-condition-safe. Two users swapping simultaneously shouldn't corrupt state.
Terraform-AWS-Cloudflare Stack Automated certificate issuance, DNS routing, and ALB ingress for Kubernetes. Built this after doing it manually once and never wanting to do it again.	Neural Network from Scratch Feedforward network in PyTorch with no nn.Linear or autograd shortcuts. Backprop, Adam, and gradient clipping done by hand. Built to understand what actually happens during training.

• Kubernetes CNI plugins and pod networking internals — I can operate a cluster, not fully explain it yet
• Distributed tracing (Jaeger/Zipkin) — I understand logs, cross-service causality is still fuzzy
• Advanced CUDA optimization — GPU memory management is where I currently get stuck

I prefer Vim over VSCode, logs over dashboards, and infrastructure as code over ClickOps.

If something breaks, I fix it. If I can't fix it, I rip it apart until I understand why it broke in the first place.

I'm open to collaborating on infrastructure-heavy ML or automation projects, things that combine scalability, reliability, and compute performance.

I value feedback on architecture and code design, especially from people who care about systems thinking.

Contributors are welcome if they want to understand, not just ship.