~/samarpit-nagpal GURUGRAM — 00:00

$ status — open to opportunities · AI Agent Engineer · Full-Stack Developer · Gurugram, IN

Samarpit Nagpal

AI Agent Engineer & Full-Stack Developer shipping production-grade systems — AI-agent pipelines, typed APIs, real-time platforms, and interfaces people actually enjoy using. Designed, built, and maintained end-to-end.

$ ls ./work — 7 projects

Selected Work

Greeto

AI Greeting-Card App · Backend & Admin Panel · Mediology

The backend and admin panel behind a consumer app on Google Play and the App Store: every greeting card is composed by a multi-model AI pipeline — text, image, and vision models orchestrated as async jobs, with per-image cost tracking, daily caps, and a batch mode that halves inference spend — plus end-to-end subscription billing driven by store webhooks.

$ react / node / postgresql / gemini · openai / job queue / store webhooks / cron

multi-model pipelineasync orchestrationcost guardrails
2026

Ad Traffic Fraud Detection

Invalid-Traffic Platform · Mediology · Internal

Fraud-detection platform reconciling Google Ad Manager and GA4 signals across 350+ publisher domains and ~100M ad impressions — scoring every domain's risk, then vetting suspects with a Playwright crawler fleet and AI research agents before revenue is on the line.

$ react / node / google ad manager · ga4 apis / playwright / gemini

risk scoringai research agentscrawler fleet
2026

Dynamic Config — SORTD

Publisher Config Middleware · Mediology

Config middleware inside SORTD, Mediology's publisher platform: publishers tune app behaviour from a UI — with an AI assistant that makes the changes for them and a live preview against the app — and updates propagate to production on the fly. No release cycle, no developer round-trip.

$ react / node / rest apis / ai assistant / live preview / on-the-fly config delivery

ai assistantzero-release updateslive preview
2026

Online Library Platform

Full-Stack · Personal Build · Live

eBook marketplace with an integrated EPUB reader — highlights, notes, themes, progress tracking — plus author publishing, JWT auth with email verification, and Stripe checkout across 30+ REST endpoints.

$ react 18 / typescript / redux-toolkit / node / express / mongodb / stripe / epubjs

Bluestar PCM

IoT Monitoring Platform · IOTFY

Real-time monitoring for 3,000+ deployed IoT devices. Live fault detection and daily monitoring workflows that cut support response times for multiple internal teams.

$ typescript / mqtt / websockets / rest apis / real-time telemetry

2025

Zeno — Factory PMS

Production Management · IOTFY

End-to-end production management across 6+ factory stages — kit planning, SMT, assembly, testing — with OK/NG/rework flows tracking every manufactured device, gated by role-specific backend APIs for 10+ roles.

$ typescript / node / postgresql / rbac / workflow engine

2025

Hotel Management System

Dual-Portal Booking System · Personal Build

Two connected apps: a server-rendered Next.js 14 guest portal with real-time availability and dynamic pricing, and a React admin dashboard with KPI analytics, revenue tracking, and full booking & cabin management.

$ next.js 14 / react query / supabase / recharts / ssr

$ whoami

About Me

I like software you can lean on — the kind teams run their day on, and now the kind where agents do the heavy lifting.

I'm Samarpit — a software developer in Gurugram, currently at Mediology Software, building AI-agent systems end-to-end: a consumer app whose greeting cards are composed by a multi-model pipeline, research agents that hunt ad fraud across hundreds of publisher domains, and the job queues, cost guardrails, and billing webhooks that keep it all shippable.

Before that, at IOTFY, I built the unglamorous, high-stakes middle of connected products: monitoring platforms watching 3,000+ devices in the field, factory floors tracking every unit from kit to test, and real-time pipelines that shaved ~80% off critical service operations.

My happy place is the full stack — a typed API on one end, an interface people actually enjoy on the other, and a database schema in between that won't embarrass me in a year. Lately that stack includes a model layer too: picking the right model for the job, keeping latency and spend honest, and designing for the runs that fail.

Off the clock: grinding data structures & algorithms, studying system design, and playing chess — where, much like in production, the losing move is always the one you didn't think through.

  • CurrentlySoftware Developer @ Mediology Software
  • PreviouslySoftware Developer @ IOTFY
  • FocusAgentic AI systems · Full-stack product engineering
  • EducationMCA, Chandigarh University (ongoing) · BCA, IGNOU
  • Based inGurugram / Delhi NCR, India
  • ElsewhereChess boards & LeetCode
$ cat ./toolbox
Languages
javascript es6+ / typescript / python / sql
AI & Agents
openai · anthropic claude · gemini apis / langchain · langgraph / agent orchestration / multi-agent workflows / tool calling · function calling / mcp (model context protocol) / rag · vector embeddings (pgvector · pinecone) / semantic search / evals · llm-as-a-judge · langsmith / prompt & cost engineering
Frontend
react / next.js / redux / react query / tailwind css / vite
Backend
node.js / express / electron / mqtt / websockets / payment webhooks / jwt · rbac
Data
mongodb / postgresql / mysql / supabase / redis
Cloud & Ops
aws lambda / s3 / ec2 / docker / playwright automation / linux / ci-cd

$ mailto — have a role, a product, or a stubborn bug?

Let's build

Usually reply within a day. Happy to walk through code, architecture decisions, agent pipelines, or why that one query was slow.