Software engineer

Ritvik

IIT Ropar, ex-Rakuten

I've spent three stretches on other people's engineering teams — BYJU'S, Rakuten, and a two-million-download learning app — and the year since as the only engineer on five different products, from idea to something a client's customers actually use. Same instinct either way: find what's actually broken, build the smallest thing that fixes it, and don't let the codebase get away from you as it grows.

BYJU'S

Rakuten

2M-download app

Independentnow

now

Sole technical owner on a handful of client engagements right now — architecture, delivery, no team behind me.
Most recent build: Runbook, an on-call incident copilot. Hybrid retrieval, an evaluation suite that gates every change, a safety layer enforced in code, and a decision record for every architectural call. Deployed.
Considering one full-time or contract seat this month, if the fit is right. Remote, async-first.

how I use AI

I use Claude Code the way I'd use a very fast, very literal junior engineer — it's excellent at producing a first draft and terrible at knowing when the first draft is wrong. My job hasn't changed, it's just moved earlier: deciding what should be built, catching the plausible-looking thing that's actually broken, and being the one who owns it when the pieces have to fit together. Every project below has a moment where the obvious AI-generated answer wasn't the right one, and I built the thing that actually was.

how I decide

Diagnose before you optimize.

The fix is usually not “add a cache” — it's finding what's actually slow first.

Keep AI away from anything where a wrong guess costs money or trust.

Everywhere else, let it move fast.

Build it custom only when reuse will actually pay for the extra work.

Otherwise, wire together what already exists.

Assume every step gets retried eventually.

Make retrying safe from day one, not after the first incident.

the work

What I've actually built

● live

Independent · Self-built, live

featured

Runbook — incident copilot

A deployed on-call copilot that handles the first fifteen minutes of an incident. It's also where I went deep on the parts that separate a real LLM system from a demo: retrieval that cites its sources, evals that gate every change, and a safety layer that lives in code, not in a prompt the model can ignore.

PythonFastAPIPostgres + pgvector (Neon)Hybrid retrieval + rerankingOpenRouter (model routing)LangfuseReactDocker / Renderpytest + CI

read the build →

Independent · Live, multi-tenant

Cold Outreach OS

A multi-tenant outreach operating system running live campaigns for multiple agency clients from one dashboard, with three lead-gen engines behind it.

Next.jsNode.js worker servicePostgres (Neon)Drizzle ORMApolloExploriumApifyPerplexityZeroBounceInstantlyOpenAI / Claude

read the build →

▶ interactive tour

Independent · Live

Social Intelligence Platform

Six-platform social listening, built so every sentiment claim traces back to a real, clickable post.

Next.jsPostgres (Neon)n8nApifyPerplexityOpenAI (GPT-4o / GPT-4o-mini)

read the build →

▶ interactive tour

Independent · Live

Amazon Listing Intelligence

Reads the top 30 competitor listings in a category — copy, reviews, photos — and writes a new listing meant to beat them.

ReactViteExpress.jsPostgresn8nApifyOpenAI (GPT-5.1, GPT-4o vision)

read the build →

▶ interactive tour

Full-time — PrashantAdvait Foundation · Live, 2M+ downloads

LMS Performance Overhaul

Took a core screen from 9.2 seconds to 141 milliseconds by finding the actual bottleneck: missing indexes and queries pulling data nobody needed.

GolangCouchbase (N1QL)Clean ArchitectureTDD

read the build →

Full-time — Rakuten Group, Tokyo · 16M+ user app

Rakuten

An internal tool that saved 200 engineering hours a year, and a battery-drain fix on a 16M-user banking app, done alongside a team at Google.

AndroidJenkins CI/CD

read the build →

Full-time, six months — core backend team · Content-recommendation service

BYJU'S

Helped take a Ruby monolith apart and rebuild pieces of it as Go microservices, live, with real users on both systems at once.

RubyGolangAWSKafkaDocker

read the build →

Independent · Self-built, live

FDA & Trial Signal Detector

Watches FDA filings, clinical trials, and grant data for signs a biotech company is about to need a vendor — and a real story about catching its own mistakes before they mattered.

PythonPostgreSQLStreamlitSEC EDGARClinicalTrials.gov APINIH RePORTER API

read the build →

more work

Five-Minute Site Agent

Give it a Google Maps link, get back a live, deployed website with its own design choices already made — logo, colors, layout — in about five minutes. Built first for local clinics in India, but the pipeline itself — scrape, decide, build, deploy — isn't specific to either the industry or the country. I also built and ran the sales motion around it myself: cold outreach, a demo-first pitch, real pricing, a real close rate.

what this isn't for

Probably not a fit if the job is built around being in meetings all day, or if “engineer” really means “implement exactly this spec, no questions.” I like having a team around — three of my last four roles were on one — but I want to own the outcome of what I build, not just the ticket.

If this looks like a fit, reach out. Happy to get specific once I know what you're actually solving for.

ritviks001@gmail.com github linkedin