Amal Zacharia

Senior Platform & Infrastructure Engineer

Self-taught. I've held 200,000+ production connections above a client-set connection floor. Solo, end to end.

About

Self-taught Senior Platform & Infrastructure Engineer, 6+ years owning distributed systems, bare-metal Kubernetes, and high-concurrency backends for high-scale community platforms and global startups. No CS degree; the systems in production are the credential.

The proof: two years keeping 200,000+ concurrent WebSocket connections above a client-set 75% connection floor, outside planned maintenance, as a solo operator, on infrastructure I built and ran myself. Separately, migrated production workloads off managed cloud onto self-hosted bare-metal Kubernetes, cutting infrastructure spend by roughly 85%.

I work best where ownership is total and the problem is hard. No unnecessary escalation, just the system, the problem, and a solution by morning.

I've spent years as the engineer clients retained long term, paid a premium to keep, and relied on when things broke at the worst possible time, as an independent consultant since 2020.

Curious how any of this actually happened? Read the full story, download my resume, or just drop me a note.

Articles

  1. We Dropped Harbor for Large Images. A Traefik Default Was the Cause.

    2026

    Large uploads failed at random while every dashboard stayed green. Traces from Cloudflare and Traefik showed each failure ending at exactly 60 seconds.

    Thumbnail for We Dropped Harbor for Large Images. A Traefik Default Was the Cause.

  2. No Database, No Queue: Splitting 100,000 Jobs With One Bit Shift

    2026

    Ten workers, 100,000 jobs, no coordinator. How one line borrowed from Discord's sharding scheme let ten workers split a big job with no database and no queue.

    Thumbnail for No Database, No Queue: Splitting 100,000 Jobs With One Bit Shift

  3. Clean Counters Don't Prove There's No Loss: How a DHCP Pool Broke Production

    2026

    A TCP connection stalled for 30 seconds every 90 seconds while every counter stayed clean. The cause was one IP address held by two devices.

    Thumbnail for Clean Counters Don't Prove There's No Loss: How a DHCP Pool Broke Production

  4. I Benchmarked eBPF Against nftables. They Only Start to Diverge After 100 Rules.

    2026

    At 1 rule, nftables was slightly faster. At 4,000 rules, my eBPF firewall had 6x the throughput. Where each side wins, and what I never measured.

    Thumbnail for I Benchmarked eBPF Against nftables. They Only Start to Diverge After 100 Rules.

Experience

  1. Apr 2020 - Present

    Delivered backend systems and infrastructure for international clients across high-scale community platforms and AI startups. Migrated production workloads from managed cloud to self-hosted k3s on bare metal, cutting infrastructure costs by roughly 85%. Owned OS provisioning, firewall hardening, and CI/CD automation using GitHub Actions and GHCR.

    • Go
    • k3s
    • Kubernetes
    • Rancher
    • Prometheus
    • Grafana
    • Linux
    • Cilium
    • Terraform
    • Ansible
    • GitHub Actions

Certifications

Story

  1. 01

    How It Started (2018)

    I was 15, looking for a clan. What I found instead was the beginning of everything.

  2. 02

    First Lines of Code (2019)

    Still on a phone, still no PC. But I wanted to build something myself, not run someone else's code.

  3. 03

    The Year Everything Accelerated (2020)

    I stopped experimenting and started building for real. Kubernetes, TypeScript, real clients, all in one year.

  4. 04

    Scale, Systems and New Clients (2021)

    A product idea, a race condition I'd never seen before, and the mental model that would become proper distributed systems architecture.