Skip to content

Atharv Dange - Full Stack Engineer & Cinephile

atharv@dange:~$neofetch
ATHARV DANGE
Full Stack Engineer & Cinephile
OS: Node.js/Web Core
Host: Pune, Maharashtra, India
Shell: React/Node.js/TypeScript

$cat about.md

I'm a Full Stack Engineer based in Pune, India. I build production systems end-to-end, from auth protocols and API design to the interfaces people actually use. Right now I'm freelancing, most recently on a booking and payments platform for MentorAID (opens in new tab), a UPSC-prep startup in Delhi.

When I want to understand how something works, I rebuild it from first principles. I've built HTTP frameworks, client-side routers, CSS engines, OIDC providers, and web rendering pipelines. Not to replace existing tools, but to understand the problems they solve at the protocol level.

Outside of code I write about what I find at blogs and watch way too many films.

$cat experience.md

Freelance2026–present
Full Stack Engineer
  • ▸Built MentorAIDConnect with a 3-person team for MentorAID, a Delhi-based UPSC-prep startup: students book one-on-one mentor sessions and buy answer-copy evaluation packages
  • ▸Designed slot booking with 20-minute payment holds that expire on read, plus Razorpay payments that snapshot price and time onto each order
Smartscripts Pvt. Ltd2024–2026
Jr Software Engineer → Software Development Team Lead
  • ▸Built and shipped production products end-to-end across the full stack
  • ▸Took ownership of the IT department: architecture decisions, client-facing project scoping, quotations, and on-time delivery
  • ▸Mentored interns and junior developers
  • ▸Conducted technical interviews for new hires
  • ▸Entered a profit-share agreement reflecting expanded ownership and responsibility over the firm's technical direction

$ls projects/ --featured

[active][product]
Kleis OIDC

Built a production-grade OpenID Connect identity provider from scratch - implemented the OAuth 2.0 Authorization Code Flow with PKCE, RS256 asymmetric JWT signing, automatic refresh token rotation with theft detection, and shipped a plug-and-play Next.js SDK for client apps.

Next.jsTypeScriptNode.js
[active][product]
Shikai

A beautifully crafted, read-only GitHub dashboard for Android. Browse repos, explore code, track your contribution streak, and share your profile - without the pressure of taking action.

ExpoReact NativeTypeScript
[active][product]
Tabi

A group trip planner that pulls the plan, the money, and the people out of the group chat and into one place. Shared itinerary with list, calendar, and map views, expense splitting with settle-up, and role-based access checked on every request. My friends and I plan our trips on it.

Next.js 16React 19TypeScript
[completed][npm package]
Reiatsu

Built a zero-dependency HTTP server framework from Node.js core modules - designed a compile-time type-safe router with parameter inference, an async onion-model middleware pipeline, and built-in rate limiting, CORS, CSRF protection, and Brotli compression. Published to npm.

TypeScriptNode.jsHTTP(Node.js Core)
[completed][product]
XML Sitemap Generator

A production-grade web crawler and sitemap discovery engine with background job processing, real-time progress streaming, and intelligent CSR/SSR detection for creating SEO-compliant XML sitemaps.

Next.js 16React 19TypeScript 6
View all projects

$cat stack.md

Languages
TypeScript
Frontend
ReactNext.jsReact NativeExpo
Backend
Node.jsPostgreSQLMongoDBPrisma
Tooling
React QueryZustandZodTurborepoOxcVitest
Protocols
OAuth 2.0OIDCRESTSSE
Infra
VercelGitRenderHostinger

$ls blogs/ --recent

Updated October 15, 2026·22 min read

A from-first-principles case study on five web rendering strategies (Static Site Generation, Server-Side Rendering, Client-Side Rendering, Incremental Static Regeneration, and Partial Pre-Rendering), built with zero frameworks using Node.js's built-in http module.

Updated October 12, 2026·15 min read

What I learned building Mekiki, my first AI agent: an npm package advisor that kept giving wrong answers. Each time, the traces showed the model doing its best with bad input, a missing date, or a rule I had written badly.

September 23, 2026·19 min read

A concept by concept walkthrough of what happens inside a modern LLM: tokenization, embeddings, RoPE, attention, induction heads, multi-head attention with GQA, the feed-forward network, residual streams, and next-token prediction. Written up after working through the mechanics until every piece made sense.

September 7, 2026·14 min read

Two slices that teach the rest of the system to read the route tree slice 8 turned into the single source of truth. Slice 9 makes the type system check every path and param at compile time. Slice 10 replaces order-dependent first-match matching with a specificity score, so the right route wins no matter how the tree is written.

Read all posts