Open to interesting problems · Gurugram, India

I build systems that stay quiet under load.

I'm Abhinav — a backend and distributed-systems engineer. I care about the unglamorous half of software: the queues, workers, and services that have to hold when everything else is on fire. Given the choice, I'll delete a system before I add one.

See the workGet in touch
event-driven by defaultdelete before you addkeep it boring
01 — Work & experience

Where I've shipped, and what it cost in latency.

SpyneSoftware EngineerCurrent
May 2025 — Present · Gurugram

Sole engineer on a human-in-the-loop QC platform that validates AI-generated car merchandising for 5,000+ dealerships across the US and Europe — 150–250K images a day. I own the architecture, the backend, and the pager.

Event-driven QC dispatch
Replaced cron-based polling — the bottleneck in the dispatch pipeline — with an SQS + Lambda design fitted to the single-consumer dispatch pattern, without standing up a Kafka consumer for a job queue.
p95 4s → 250ms
Multi-level pipeline orchestration
Built pause/resume that halts work at SKU or full-tenant scope — blocking in-flight processing and new ingestion — and recomputes SLA timers on resume. Bad client inputs no longer force full-batch reprocessing.
< 300ms e2e
Standalone 3D asset validation
Pulled glb/ply radius, height, and background-view checks out of a legacy service that wrote validation state straight onto live customer assets, into an isolated Nest.js service with its own database.
incidents removed
Real-time branding re-evaluation
Added Go/Gin real-time re-evaluation to a rule-based branding service across ~50K live VINs — consuming the inventory-update Kafka topic, and pre-computing each VIN's next transition date to drive an EventBridge scheduler for time-based flips that emit no event.
~50K VINs
SuperjoinSoftware Engineer
Feb 2024 — Apr 2025 · Bengaluru

Built the API integration engine that grew to roughly 80% of daily user activity, and cut per-integration go-live from about four weeks to three days.

Config-driven integration framework
OAuth 2.0, Bearer, Basic auth, token refresh, and per-source rate limiting — shipped 12+ pre-packaged integrations including Google Ads, Salesforce, and Stripe, each bundling multiple entities under one config and replacing hand-written per-source code.
4 wk → 3 days
White-label platform layer
Per-partner OAuth apps, isolated deployments, and runtime branding injection on the Apps Script platform — letting partners resell the product under their own brand.
per-partner OAuth
Extractor + LLM table ranker
Replaced 15+ hand-written site-specific table parsers in the Chrome extension with a general extractor and an LLM-based ranker that surfaces the most relevant table — removing per-site maintenance and working on sites with no parser at all.
15+ parsers cut
02 — Open source

Things I build to understand things.

Go
redis-go

A Redis server reimplemented in Go — the RESP protocol, core commands, and the event loop, built to learn the internals from the ground up.

View on GitHub →
Go
rssaggregator

Scrapes posts from RSS feeds for the topics a user follows, fanning the work out concurrently with goroutines.

View on GitHub →
JavaScript
ci-cd-sample

A reference deployment pipeline — GitHub Actions building and shipping a container to AWS EC2 via Docker.

View on GitHub →
All repositories on GitHub →
03 — Writing

Notes on the systems I build.

Mostly write-ups of problems I hit at work — queues, latency, and the occasional rewrite that was worth it.

Systems
Jun 2025
Killing cron: moving a dispatch pipeline to SQS + Lambda

Why polling was the bottleneck, and how an event-driven rewrite took p95 assignment latency from 4 seconds to 250 milliseconds.

8 min
Architecture
Apr 2025
Designing pause and resume into a multi-tenant pipeline

Halting work at SKU or tenant scope without reprocessing whole batches — and getting SLA timers to survive the pause.

6 min
Notes
Jan 2025
Replacing 15 hand-written parsers with one LLM ranker

A general table extractor plus a relevance ranker that ended per-site maintenance in a Chrome extension.

5 min
04 — About

A bit about me.

I'm a software engineer in Gurugram, India. I work on the parts of a product most people never see — the queues, the workers, the services that have to keep running when traffic spikes and a model starts returning nonsense at 3am.

I like problems where correctness and latency both matter, where the right move is often to delete a system rather than add one. Lately that's meant a lot of event-driven design, a fair amount of Go, and more time than I expected thinking about how to pause a pipeline safely.

Outside the day job I rebuild tools to understand them — a Redis server in Go, a small RSS aggregator — and write up whatever I learned.

EDU
B.Tech, Computer Science & Engineering
Jaypee Institute of Information Technology · 2024
AS
Portrait
Gurugram · India
05 — Stack

Tools I reach for.

Languages
GoTypeScriptPythonC++
Backend
Node.jsExpressNest.jsGinFastAPI
Data
PostgreSQLMySQLMongoDBRedisApache Kafka
Cloud & ops
AWS · SQS · Lambda · EventBridgeECS · EC2 · CloudWatchDockerJenkinsGrafana
AI tooling
Claude CodeCursorMCP servers
NOW

Most days that's queues, workers, and keeping p95 latency boring.

Currently the sole engineer on Spyne's QC platform — a human-in-the-loop system validating AI-generated car photos for 5,000+ US and European dealerships. Between shipping, I'm writing about the systems behind it and slowly rebuilding Redis in Go for fun.

06 — Contact

Let's talk.

Have a problem worth solving, or just want to compare notes on queues and latency? Send a line.

Currently at SpyneOpen to interesting backend and infra problems.