Skip to content
Rishi YadavNarnaul, India

For recruiters and founders

Full-stack engineer working in Next.js, Node.js and applied AI

I build production websites and web apps end to end: TypeScript frontends, Node.js and Python APIs, PostgreSQL, and LLM features grounded in real data. Below is the stack I use, how I think about a few real projects, and code you can read.

Open to full-time and contract roles in full-stack and AI engineering.

Resume (PDF)GitHubLinkedInphotonics54yadav@gmail.com
Based in
Narnaul, Haryana, India. Remote across India.
Education
B.Tech, Computer Science (AI & ML), Ch. Ranbir Singh State Institute of Engineering & Technology, 2023–2027
Focus
Full-stack TypeScript, APIs and databases, LLM integrations, web performance
Live work
17 projects with live links

Stack

What I work with

Grouped by where it sits in a system. I pick the simplest tool that fits the job, and I'm comfortable reading and changing code in all of these.

Frontend

  • TypeScript
  • React
  • Next.js
  • Tailwind CSS
  • React Native
  • GSAP
  • Framer Motion

Backend

  • Node.js
  • Express
  • tRPC
  • Hono
  • Python
  • FastAPI
  • Zod

Data

  • PostgreSQL
  • MongoDB
  • Supabase
  • Prisma
  • Drizzle
  • Qdrant

AI / LLM

  • OpenAI API
  • Llama 3
  • RAG with vector databases
  • LangChain

DevOps & tooling

  • Vercel
  • AWS
  • Docker
  • Cloudflare
  • GitHub Actions
  • Shopify
  • WordPress
  • Lighthouse

Architecture notes

How a few real projects are put together

Short notes on what I built, why it's shaped that way, and what it costs. Each one links to the live project.

ReviewBoost AI: grounded AI replies

ReviewBoost AI case study
What
A FastAPI service retrieves business-specific context from a Qdrant vector store and passes it to Llama 3 to draft a review reply, which a person approves in a Next.js dashboard.
Why
Generic LLM replies read as canned. Grounding each reply in the business's own details (services, hours, tone) keeps responses specific without fine-tuning a model per customer.
Trade-offs
Retrieval adds latency and an extra service to run, but it keeps per-business data editable and avoids retraining. A human approval step trades full automation for fewer embarrassing replies.

A static-first site for an eye hospital

Goyal Netra case study
What
Treatment, doctor and contact pages are pre-rendered with Next.js and served as static HTML from a CDN. Content sits in typed files, and every page ends with the same enquiry options: call, WhatsApp or a short form.
Why
Patients mostly arrive on phones, often on patchy mobile data, and want one thing: to book. Static pages load fast without a server round trip, and repeating the same enquiry options means nobody has to hunt for a phone number.
Trade-offs
Content changes need a rebuild and deploy instead of a CMS edit. For a site that changes a few times a month that's acceptable, and it removes a database, an admin login and a whole class of security updates.

Large photos without a slow page

Chandni Resorts case study
What
Room, venue and event photos go through next/image: responsive sizes per breakpoint, modern formats (AVIF/WebP) where the browser supports them, explicit width and height, and lazy loading for everything below the fold. Only the hero image loads eagerly.
Why
A resort sells with photos, but uncompressed galleries are the most common reason hospitality sites feel slow. Serving the right size to each device keeps the photos sharp without making a phone download desktop-sized files.
Trade-offs
On-demand image optimisation uses hosting compute and adds a short delay the first time each size is requested. Pre-sizing images by hand would avoid that but makes every photo update a chore for the client.

Routing enquiries to where the team already works

Pariksha Career Institute case study
What
Enquiry forms validate on the server, drop obvious spam with a honeypot field, and send the details by email. Next to each form, a WhatsApp link opens a chat with a pre-filled message so people can skip the form entirely.
Why
Small teams at institutes and clinics reply on WhatsApp and check email, not a dashboard. Sending leads straight there means enquiries get answered the same day instead of sitting in a database nobody opens.
Trade-offs
There's no CRM, so reporting is limited to counting emails and analytics events. That's the right trade for a team of a few people; a bigger team would want enquiries stored and assigned.

GitHub

Public repositories

A few recent public projects. Everything else is on github.com/rishiyadav11.

Want to talk about a role?

Email me with the role, team and stack. I reply within one working day.

Email meView resume