Backend • AI • Cloud Engineer | Builder of agentic systems | Currently learning Go

I’m Karthik Nadar, a backend, AI, and cloud engineer who loves building systems that actually work at scale.

I started coding in my second year of college, right after almost being ousted in the first round of my first-ever hackathon. That moment stung — but it lit a fire in me to never be underprepared again.

I began with React and JavaScript, building small projects that taught me how things worked under the hood. Soon after, I joined GirlScript Summer of Code, a local open-source initiative, where I ranked 22nd among 1000+ contributors. That was my first milestone — proof that persistence compounds.

From there, I went deeper into backend engineering with Express.js and FastAPI, built full-stack apps, and learned the hard lessons of debugging, scaling, and deployment. Around this time, I stumbled into AI — and it changed everything.

My first real AI project was a small-scale clone of Perplexity’s search system. I sketched the entire architecture on my notebook during my microprocessor class and then in my dorm wardrobe — from the data flow to the agentic workflow — and then brought it to life in code. That was my “lightbulb” moment: realizing I could build something end-to-end that thinks.

Later, I competed in HackRx 5.0, hosted at Bajaj Finserv HQ (Pune), where I was among the top 22 teams across India. We built an AI video generator using Next.js, Python, and Remotion, another project that started out with an architecture drawing on my wardrobeto builting multiple iterations to see what works and what doesn't.

Soon after, at HackByte 3.0 (IIIT Jabalpur), I built DataShorts — a platform that lets you converse with your database in natural language. Competing among 300+ teams, we placed 4th nationwide, and the project was praised for its chunking strategy, retrieval design, and clean architecture.

That hackathon project is now evolving into my personal venture — datashorts.com — where I’m building an intelligent interface between humans and structured data.

I believe that the next leap in AI won’t come from bigger models — but from agents that remember.

I’ve been diving deep into AI memory systems and personalization, inspired by the Mem0 research paper, experimenting with:

• Vector memory upsertion and relevance scoring
• Tool-calling for selective long-term retention
• Graph-based memory relationships
• Context summarization pipelines

These experiments aim to answer one question:

“How can we make AI remember like humans — contextually, intelligently, and efficiently?”

One of my favorite learning methods is to reverse-engineer the tools I admire.

I’ve built smaller-scale clones of:

• LangSmith → to understand how LLM tracing and observability platforms work,made me realise how there is so much more to ai than just calling apis and models.
• Perplexity → for retrieval, contextual Q&A, and agentic orchestration,my first intro to agentic systems and how memory is maintained across agents.
• AWS S3 wrapper → for simplified cloud storage APIs,super exciting to actually get over the complex s3 setup and build something that does the same in a single click.

These projects taught me why great systems feel simple— they hide the complexity behind clarity.

Chat with your database in natural language. Transform structured and unstructured data into actionable insights instantly. Built using nextjs,postgresql,pinecone vector db and openai api for intelligent query generation and simulation in real time. in the middle of migrating to express server and add a graph db layer and some fuzzy matching logic for mispelled terms so as to not waste user tokens.

AI-powered search assistant inspired by Perplexity. Built with OpenAI, Tavily Search API, and Vercel AI SDK for contextual, retrieval-augmented responses.

Smart note-taking app that lets you chat with your notes. Built with OpenAI api, all notes get converted to embeddings in real time and a modal to chat with your notes in real time.

Kyra sdk

A lightweight LangSmith-inspired observability SDK for LLM applications. Built from scratch to understand the inner workings of LLM tracing, evaluation, and session-level analytics. Kyra helps log model interactions, visualize traces, and analyze reasoning patterns for agentic workflows. This was my deep dive into AI infrastructure, tracing, and developer experience — reverse-engineering the principles behind LangSmith to truly understand how observability in AI works.

• 22nd out of 1000+ — GirlScript Summer of Code 2023
• HackRx 5.0 (Bajaj Finserv HQ, Pune) — Top 22 teams nationwide
• HackByte 3.0 (IIIT Jabalpur) — 4th out of 300+ teams

Let’s connect, collaborate, or brainstorm something fun:

• LinkedIn
• Hashnode
• Twitter
• Instagram
• GitHub

Thanks for stopping by — I’m always happy to connect, collaborate, or just talk about AI systems, Golang, and scalable software engineering. Let’s build something that actually works — and remembers