# Nikhil Thakuraan — Freelance Full-Stack & AI Developer (Jaipur, India) Freelance full-stack and AI developer based in Jaipur, India. Builds modern React / Next.js websites and landing pages for small businesses, creators, and early-stage startups (recent example: sarinacheesecake.in), and ships AI work — OpenAI and Claude API integrations, RAG pipelines, LangChain apps, custom AI chatbots, and AI agents. Also publishes original AI tooling including the poemformer transformer model on Hugging Face and the sedimem layered memory system for LLM agents. Available for hire worldwide. Contact: contact@nikhilthakuraan.in. Keywords: freelance AI developer Jaipur, freelance web developer Jaipur, freelance full stack developer Jaipur, AI freelancer Jaipur, LangChain developer Jaipur, OpenAI integration developer India, Claude API developer India, RAG developer, AI chatbot developer Jaipur, Next.js developer Jaipur, React developer Jaipur, MERN stack freelancer, SaaS developer Jaipur, AI agents, AI automation, vector database, pgvector, Pinecone. --- ## About I'm Nikhil, 18, based in Jaipur, on a gap year, mostly building things on the internet and lifting heavy things. I freelance — modern websites and landing pages (React, Next.js, Tailwind) for small businesses, creators, and early-stage startups across India and abroad. I also ship AI work for clients: OpenAI and Anthropic / Claude API integrations, retrieval-augmented generation (RAG) pipelines, LangChain apps, custom chatbots, AI agents, and small SaaS tools. Stack on that side is mostly Python and Node with pgvector / Pinecone-style retrieval. On the research side I trained a small poetry transformer (poemformer, published on Hugging Face) and keep building layered memory systems for LLM agents (sedimem, recall, hippocampus) because the forgetting problem annoys me. When I'm not at a keyboard I'm reading (big Dostoevsky guy), cycling around the city, or somewhere on a train pretending to be a backpacker. ## Hire me I'm available for: - **Freelance websites & landing pages** — React, Next.js, Vite, Tailwind CSS, Framer Motion, occasional Three.js. Small-business sites, brand sites, portfolio sites, marketing pages. - **OpenAI & Claude API integrations** — adding LLM features to existing products: streaming chat, tool use / function calling, prompt caching, basic eval and monitoring. - **RAG & LangChain apps** — retrieval-augmented Q&A over docs, knowledge bases, or product data. LangChain or a thin custom layer, plus pgvector or Pinecone for vector search. - **Custom AI chatbots & agents** — domain-specific chatbots, internal copilots, lightweight multi-step agent workflows. - **AI automation & internal tools** — AI-assisted email triage, document Q&A, lead enrichment, custom scrapers, FastAPI / Next.js back-ends. - **Small SaaS MVPs** — full-stack JavaScript / TypeScript, Postgres, Vercel / Render deploys. - **AI tooling collaborations** — especially memory and retrieval systems for agents. Email: contact@nikhilthakuraan.in. ## Work ### Freelance web & AI development I've shipped websites, landing pages, brand refreshes, and AI integrations for multiple small-business, creator, and early-stage startup clients across India and abroad. Stack: React, Next.js, Vite, Tailwind CSS, Framer Motion, Three.js, Node.js, TypeScript, Python, OpenAI API, Anthropic Claude API, LangChain, pgvector, Pinecone, Postgres, FastAPI, Vercel, Render. Recent example client: - **Sarina Cheesecake** (https://www.sarinacheesecake.in/) — bakery website, full design and build in React + Tailwind, shipped in a week. ### AI work - **AI chatbots & RAG apps (freelance)** — custom chatbots and retrieval-augmented apps for clients. OpenAI / Claude integrations, LangChain orchestration, vector search with pgvector or Pinecone, deployed on Vercel / Render. - **poemformer** (https://huggingface.co/rororova/poemformer) — my own transformer model fine-tuned for generating poetry. Trained from scratch in Python / PyTorch and published on Hugging Face. - **sedimem** — a layered memory system for LLM agents (recent / compressed / archived tiers instead of flat embedding soup). - **recall / hippocampus** — durable memory and project-context tooling for coding agents. Vector retrieval, knowledge graph integration, persistent notes. ## Profiles - GitHub: https://github.com/Rororova - X / Twitter: https://x.com/Nikhilthakuraan - LinkedIn: https://www.linkedin.com/in/nikhil-thakuraan/ - Hugging Face: https://huggingface.co/rororova ## Articles ### Why I took a gap year Gap year — read: a long, weird, mostly unsupervised lab. Everyone around me went straight into college. I didn't. Not because I was lost — because I wanted to actually figure out what I'd be building for the next 10 years before I let a syllabus pick it for me. So far it's been: ship stuff, read a lot, lift heavy, talk to interesting people on the internet. I'm learning faster than I ever did in a classroom. If you're on the fence about a gap year — do it, but have a plan. Unstructured time is a gift and a trap. ### Getting paid to build websites at 18 First freelance client paid me before I owned a real desk. The trick wasn't being the best designer or the cleanest coder. It was: show up, listen harder than they expect, and ship something they can show their friends in a week. Small businesses don't want a 30-page strategy doc. They want a site that loads fast, looks like them, and doesn't embarrass them when their cousin visits the url. sarinacheesecake.in was one of those. ### Why I keep building memory systems for AI LLMs forget. That's the whole bug. You can have the smartest model in the world and it'll still ask you what your stack is on turn 12 of the same conversation. So I keep ending up in the same problem space: how do you give a model a real, durable memory without turning it into a vector-search soup. That's what sedimem is about — memory that settles into layers (recent vs. compressed vs. archived) the way human memory does, instead of flat embeddings everywhere. The next 2 years of AI tooling are going to be less about bigger models and more about boring infrastructure like this. Memory, retrieval, scheduling. Boring is where the money lives. ### Building from Jaipur, not Bangalore Everyone in Indian tech says you have to be in Bangalore. I'm not. I'm in Jaipur and I think that's quietly an edge. The rent is cheap. The food is better. Nobody is networking at me at a co-working space at 8pm. I have time to think. The internet is the same everywhere. The people I want to talk to are mostly on it anyway. And when I do need a city, it's a 4-hour train to Delhi. You don't have to be in the obvious place. You just have to actually ship. ### The case for cold DMs Most of the good things in my life this year came from messaging a stranger who had no reason to reply. Rules I follow: 1) Be specific — "I liked your work" is noise, "I rebuilt your landing page hero, here it is" is signal. 2) Be short — if it doesn't fit in their preview pane, it doesn't get opened. 3) Don't ask for anything on the first message. People massively overestimate how rude cold DMs are. What they actually are is rare and useful, if you do them with a little taste. ### On AI hype (a 2026 take) Half the internet thinks AGI is here. The other half thinks LLMs are a parlor trick. Both are wrong in boring ways. What's actually happening: models got good enough that the bottleneck moved. It used to be "can the model do the task". Now it's "can we give it the right context at the right time". That's a tooling problem, not an intelligence problem. Which is why I find the unsexy parts more interesting — memory, retrieval, scheduling, eval. The model is the engine. Nobody got rich selling engines without a car around them. ### Why I write on the internet I write because it's the cheapest way to find out what I actually think. Half my opinions don't survive the second paragraph. That's the point. Also: writing in public is a slow-motion magnet. The right people find you over months and years, not minutes. You just have to keep being findable. ### Gym at 6am is a cheat code I lift before most of the city is awake. Not because I'm a morning person — I'm not — but because the version of me that shows up at 6am has no excuses left by 9am. The work is already done. Everything after is bonus. Also: cold gym, empty rack, no waiting. You can't beat the math. ### Why I keep coming back to Dostoevsky People think Dostoevsky is heavy. He is. But he also wrote the most honest characters in literature — people who contradict themselves mid-sentence, the way real people actually do. Notes from Underground is 150 pages of a guy arguing with himself and losing. It is also one of the most useful books I have ever read, because once you see that pattern you start spotting it in yourself. Read the Russians. They're funnier than you think. ## FAQ **Who is Nikhil Thakuraan?** Nikhil Thakuraan is an 18-year-old freelance full-stack and AI developer based in Jaipur, India. He builds React / Next.js websites and landing pages for small businesses, ships OpenAI and Claude integrations, RAG and LangChain apps, custom chatbots and AI agents for clients, and works on AI memory research including the poemformer transformer model on Hugging Face and sedimem. **Is Nikhil Thakuraan available for hire?** Yes. He is a freelance full-stack and AI developer available for: React / Next.js websites and landing pages, OpenAI and Claude API integrations, RAG pipelines, LangChain apps, custom AI chatbots, AI agents, and small SaaS builds. Based in Jaipur, works with clients worldwide. Reach him at contact@nikhilthakuraan.in. **What kind of websites does Nikhil build?** Modern, fast small-business websites, landing pages, and brand sites — primarily React and Next.js with Tailwind CSS, Framer Motion, and occasional Three.js. Recent example: sarinacheesecake.in. **Does Nikhil build AI chatbots and RAG apps?** Yes. He builds custom AI chatbots, retrieval-augmented generation (RAG) pipelines, and LangChain-based apps for freelance clients. Typical stack is OpenAI or Claude as the model, LangChain or a thin custom layer for orchestration, pgvector or Pinecone for vector search, and Next.js or Node for the API and UI. **Can Nikhil integrate OpenAI or Claude into an existing product?** Yes. He has shipped OpenAI (GPT-4 / GPT-4o) and Anthropic Claude integrations into client web apps — including streaming responses, tool use / function calling, prompt caching, and basic eval / monitoring. He works comfortably in Python and Node. **Does Nikhil work with MERN / Next.js / full-stack JavaScript?** Yes. He builds full-stack apps in the React / Next.js / Node ecosystem, and is comfortable with TypeScript, Tailwind, REST and tRPC APIs, Postgres (often with pgvector), and Vercel / Render deployments. **Does Nikhil do AI automation or internal tools for businesses?** Yes. He builds small AI automations and internal tools — AI-assisted email triage, document Q&A, lead enrichment, custom scrapers, and lightweight agent workflows. Usually shipped as a small Next.js or FastAPI app with an LLM behind it. **What is poemformer?** poemformer is a transformer model Nikhil trained and fine-tuned for generating poetry. It is published on Hugging Face at https://huggingface.co/rororova/poemformer. **What is sedimem?** sedimem is a layered memory system for LLM agents that Nikhil is building. Instead of flat vector embeddings, sedimem organizes memory into recent, compressed, and archived tiers — closer to how human memory actually works. **Where is Nikhil based?** Jaipur, Rajasthan, India. He works with freelance clients worldwide and is happy to take async, remote engagements. **Where can I find Nikhil online?** GitHub: https://github.com/Rororova X: https://x.com/Nikhilthakuraan LinkedIn: https://www.linkedin.com/in/nikhil-thakuraan Hugging Face: https://huggingface.co/rororova Email: contact@nikhilthakuraan.in