Software that scales.Intelligence that ships.
I'm Shaurya — a Software Engineer bridging product and infrastructure. I build backend, AI, and full-stack systems that run in production.
Ibuildbackend,AIandfull-stacksystemsthatruninproduction.FromLLMevaluationpipelinesthatscore76K+submissions,toKubernetesmigrationsthatliftuptimeonmillion-userplatforms,toRedislayersthatcutlatencyby93%.Icareabouttheboringparts—atomictransactions,exactly-onceprocessing,deterministicscoring—becausethatiswherescaleactuallylives.
Numbers that ship in production.
Real systems, real users, measurable outcomes.
Where I've built.
4+ years shipping production systems end-to-end.
Rencom Networks
Designed and shipped a 4-stage GPT-4o-mini evaluation pipeline scoring submissions for plagiarism risk, quality, factual claims, and AI-authorship — saving ~8,800+ review hours across 76,000+ submissions.
Refactored synchronous AI evaluation into an async job queue with atomic DB transactions, cutting wait time from 3–4 minutes to near-instant with exactly-once processing.
Grew the publishing platform to 575,000+ article views and 500+ approved contributors.
Shipped a tier-based contributor incentive system that curbed farming abuse while retaining high-quality contributors.
iTinker
Containerized and migrated four legacy PHP applications from EC2 to Dockerized Kubernetes on AWS EKS — raising uptime on a 1M+ user platform from 85% to 99%.
Replaced deprecated APCu with a shared Redis cluster, cutting leaderboard response times from 3s → 200ms for 100K+ concurrent users.
Built Jenkins CI/CD pipelines for Docker/K8s deployments, cutting release time from 2 hours to 15 minutes.
Built Next.js frontend + Lerna monorepo sharing API modules between React web and React Native apps.
Things I've built.
From LLM pipelines to Kubernetes migrations — production-grade systems.
TrueBearing AI
A retrieval-augmented generation system over federal immigration rules with a deterministic CRS scoring engine. Auditable results by design — exact scoring stays separate from LLM inference.
LLM Evaluation Pipeline
End-to-end pipeline that scores every submission for plagiarism, content quality, factual claims, and AI authorship — replacing a manual editorial bottleneck at scale.
K8s Migration @ 1M users
Eliminated version-conflict outages and enabled instant rollbacks. Redis cluster replaced APCu, unlocking horizontal scale across pods.
Realtime WebSocket Feed
Replaced per-client polling with a single batched WebSocket feed serving realtime data to web + mobile — reducing third-party API spend as clients grew.
How it actually works.
Deep dives into the systems behind the projects.
What I work with.
A pragmatic stack — chosen for scale, reliability, and developer velocity.
Let's build
something great.
Open to Canada & remote opportunities. I reply within a day.