I’m Harsh N. Chandak,
• M.S. Computer Science @ ASU (4.0)
• Full-stack engineer: product & AI systems
• Ships fast, scales clean & breaks nothing (mostly)
(My name means "Happiness" in Hindi. Yes, I'm aware of the irony.)
"Good code is like a good joke: if you have to explain it, you messed up."– Harsh N. Chandak
// About Meconst dev = {
name: "Harsh N. Chandak",
title: "Full-stack Developer | AI Engineer | Research Assistant",
education: "MSCS @ASU",
philosophy: "Ship if chaotic, automate if boring, scale before needed & have AI do the boring heavy lifting",
core_skills: [
"Full-stack Development",
"AI Systems & LLM Pipelines",
"System Optimization",
"System Design",
"Telemetry & Analytics"
],
traits: [
"system-thinker",
"api-gardener",
"legacy-fixer"],
daily_setup: [
"☕ Coffee",
"Postman",
"VS Code",
"GPT & Claude",
"too many browser tabs"
],
tech_stack: [
"JavaScript",
"TypeScript",
"Python",
"React",
"Next.js",
"Node.js",
"Express.js",
"FastAPI",
"PostgreSQL",
"MongoDB",
"Redis",
"Docker",
"Kubernetes",
"AWS",
"Kafka",
"Git",
"MySQL"]};
// Production system built at Arizona State University.
Built a multi-stage AI pipeline integrating audio diarization, LLM refinement, and automated image generation with rubric-based quality scoring.
// Internal developer tooling built at Arizona State University.
Designed and shipped a full telemetry system from scratch — event schema, ingestion service, storage model, and analytics dashboard.
// A personal project born from job hunt frustration.
Used Puppeteer + cron + MongoDB with JWT-auth to personalize tracking and statuses. Saving hours of weekly job hunt efforts for real-time discovery.
// ASU course project on distributed systems.
Built a Kafka → Neo4j stream processor with Dockerized BFS/PageRank analytics. Enabled live user graph insights with real-time updates under load.
// ASU big data and geospatial analytics course.
Analyzed large spatial datasets via Spark + Scala + Spatial SQL. Optimized joins, geohash clusters, and RDDs for fast geospatial queries.
// ASU D3.js data storytelling project.
Mapped 185K+ accident records using D3.js & GeoJSON. Built six interactive charts, placing 3rd in class data storytelling showcase.
// ASU Knowledge Representation and Reasoning project.
Simulated warehouse robot logic using Clingo + Python wrapper. Solved multi-agent delivery plans with optimal horizon search in ASP.
Built pipelines that don't just run models, they question them.
Evaluation loops across 10K+ inputs to catch failure modes before users do.
Rubric-based scoring with thresholds, retry logic, and deterministic selection.
️ WhisperX diarization to LLM refinement to image generation, end to end.
Because shipping AI without evaluation is just vibes with a GPU.
Taming Complexity and operations, from sales to dispatch.
️ Built inventory logic that auto-rebalanced stock across warehouses, yes, it had opinions.
Designed BI dashboards that told stories, not just numbers (hello, exec summaries).
️ Production planning? Gantt-inspired timelines that respected both dev cycles and machine breakdowns.
Assumed failure as the default, built like it.
If you gave me your ops chaos today, I’d model it by Monday and stress test it by Friday.
It isn't “build a store.” It is: make commerce sing.
Shipped 15+ full-stack storefronts. From fashion, accessories, to hardware and spare parts.
Integrated payments (Razorpay/Stripe) with grace, retries, refunds, and regrets included.
Backend: custom schemas per vertical. Frontend: fast enough to feel native.
Dashboards that helped sellers sell, not confuse them.
Juggled orders, stock, delivery, roles, and edge cases without dropping the vibe.
Real-time isn't speed, it's trust. And I build like I mean it.
Location tracking with socket + fallback polling, accurate and battery-friendly.
Chat apps with sticky state, retry queues, and graceful reconnects.
Built for poor networks. Always. Because people use phones, not fiber.
UI that reacts fast but never freaks out. Real-time should feel smooth, not sweaty.
Debugged packet drops with browser tabs and sheer will.