Abeer Choubey

Backend Engineer · Systems Architect

// Based in India.  ·  Available for opportunities

↓ scroll

The systems behind the products you trust were built by engineers like me.

I design infrastructure that scales, backends that hold, and systems that don't wake their owners up at 3am. The invisible layer — made deliberately.

How I think about systems

CLIENT Browser · Mobile · API Consumer EDGE CDN · Load Balancer · Rate Limiter APPLICATION REST API · Auth · Business Logic · Background Jobs PostgreSQL Source of truth Redis Cache Speed layer Message Queue Async dispatch INFRASTRUCTURE Docker · Linux · Networking · DNS OBSERVABILITY Logs · Metrics · Traces · Alerts

// This architecture pattern underlies most of the projects below

#01

ARK CRM

Cold-pressed oil business — full-stack CRM

Built for a solopreneur who needed enterprise CRM without the enterprise overhead. React frontend, Supabase backend. The architecture challenge was making complex relational data feel simple to a non-technical user without sacrificing data integrity.

React UI Supabase Auth PostgreSQL Row Level Security
-- ARK CRM core schema TABLE customers ( id uuid PRIMARY KEY, name text NOT NULL, phone text, tier enum('standard', 'wholesale'), created_at timestamptz DEFAULT now() ); TABLE orders ( id uuid PRIMARY KEY, customer_id uuid REFERENCES customers, total_ml integer, status enum('pending', 'fulfilled'), created_at timestamptz DEFAULT now() );
#02

Veylox

Premium digital agency — performance-first frontend

An agency landing page built with the care of a product. Vite + React. The design challenge was creating a premium feel without over-engineering the build. Result: sub-1s FCP, Lighthouse 97+, zero runtime dependencies in production.

veylox.abeer.lol ↗ Vite Build React Components Static CDN
// Veylox performance targets FCP: < 0.9s LCP: < 1.4s CLS: 0.000 TTI: < 1.8s Build output: dist/ 44 kB (gzipped) chunks: 3 (vendor split) images: WebP, lazy-loaded
#03

AI Compositor

Compositional AI workflow engine

An AI system that chains model calls into coherent multi-step workflows. The engineering challenge was managing state between inference calls without a traditional server — using event-driven architecture with a queue-backed job processor.

Input Queue LLM Calls State M. Output
// AI Compositor flow STEP 1: ingest(input) → validate schema → enqueue(job_id, payload) STEP 2: process(job_id) → fetch context from cache → call_model(prompt, context) → write result → job_store STEP 3: compose(results[]) → merge outputs → validate coherence → emit final artifact
#04

GymLedger

Personal training tracker — built like a product

A fitness tracking application built not because similar apps don't exist, but because building it right was the point. PostgreSQL stores every workout as an append-only log. The schema was designed for time-series queries before the first line of application code.

Input UI API Layer PostgreSQL time-series schema
-- GymLedger: append-only workout log TABLE workout_sessions ( id uuid PRIMARY KEY, started_at timestamptz NOT NULL, ended_at timestamptz, notes text ); TABLE exercise_sets ( id uuid PRIMARY KEY, session_id uuid REFERENCES workout_sessions, exercise text NOT NULL, reps smallint, weight_kg numeric(5,2), logged_at timestamptz DEFAULT now() );
[ THE LAB ]

The systems I build for myself are the ones that taught me everything.

HARDWARE Primary Server x86 · 16GB RAM · 2TB Pi Cluster Pi 4 ×3 · ARM · Kubernetes Network Switch Gigabit · VLAN-capable SERVICES Gitea Self-hosted Git Portainer Container Mgmt Uptime Kuma Monitoring PostgreSQL Database Nginx Proxy Reverse Proxy Wireguard VPN ACCESS Remote SSH Terminal access Tailscale Mesh network Cloudflare Tunnel Public exposure
UPTIME 99.4%
SERVICES 12 running
STORAGE 1.4 TB / 2 TB
LAST BOOT 47d ago

The homelab is where the abstractions disappear. You can't pretend to understand networking until you've debugged your own DNS at midnight. Every production decision I make is informed by something I broke here first.

[ HOW I BUILD ]
on architecture

Design for failure. Everything else is a feature.

on complexity

The simplest system that solves the real problem wins.

on ownership

If it breaks at 3am, I should be the one who built it.

on documentation

Undocumented systems are gifts to your future enemy.

[ CONNECT ]

Let's build something
that matters.

Open to senior engineering roles, infrastructure consulting, and genuinely interesting problems.