DIKSHIT

FOUNDER

Full-stack and agentic AI developer. First year at NITK Surathkal.

I'm seventeen. I spent two years preparing for JEE and missed the college I wanted. This site is what I did next.

Taking client projects through Meritverse

DRJ

DIKSHIT RJ // NITK MME EST. 2008 // BENGALURU

ID: 730-BRICK-CORE

AVAILABLE FOR ADVISORY & AUTONOMOUS AI AGENT SYSTEMS // NITK SURATHKAL // MERITVERSE PVT. LTD. AVAILABLE FOR ADVISORY & AUTONOMOUS AI AGENT SYSTEMS // NITK SURATHKAL // MERITVERSE PVT. LTD. AVAILABLE FOR ADVISORY & AUTONOMOUS AI AGENT SYSTEMS // NITK SURATHKAL // MERITVERSE PVT. LTD. AVAILABLE FOR ADVISORY & AUTONOMOUS AI AGENT SYSTEMS // NITK SURATHKAL // MERITVERSE PVT. LTD. AVAILABLE FOR ADVISORY & AUTONOMOUS AI AGENT SYSTEMS // NITK SURATHKAL // MERITVERSE PVT. LTD. AVAILABLE FOR ADVISORY & AUTONOMOUS AI AGENT SYSTEMS // NITK SURATHKAL // MERITVERSE PVT. LTD. AVAILABLE FOR ADVISORY & AUTONOMOUS AI AGENT SYSTEMS // NITK SURATHKAL // MERITVERSE PVT. LTD. AVAILABLE FOR ADVISORY & AUTONOMOUS AI AGENT SYSTEMS // NITK SURATHKAL // MERITVERSE PVT. LTD.

CHAPTER 02 // SIDE 1 // 730 DAYS

TWO YEARS.

WALL CORE SPEC: 730 COURSES SCALE: 1:1 DAY

113

STAMP 1: "Mock test. Again."STAMP 2: "Revise."STAMP 3: "One more chapter."STAMP 4: "Not yet."

730

days of prep, roughly

RECORD ENTRY // THE FOUNDATION

For two years I did one thing. I studied for JEE.

It was most of every day, for roughly 730 days. I wanted IIT. So did everyone around me, and the wanting got loud.

No single day felt like much. A chapter finished, a mock test, another revision. I only saw the shape of it later.

23k // JEE ADVANCED RANK

MISSING BRICK

CHAPTER 03 // SILENCE // 23,000

The result came. My rank in JEE Advanced was around 23,000.

IIT was out. I'd like to say I took it well. I was heartbroken. Two years, and the number that came back wasn't the one I needed.

I'm not going to dress that up.

CHAPTER 04 // SIDE 2 // FIRST LIGHT

IS THERE ANYBODY OUT THERE?

98.9

JEE Main Percentile

COLLEGE OF WEIGHT // SURATHKAL

Eventually I looked at what I actually had. A 98.9 in JEE Main. A seat at NITK Surathkal.

That isn't a consolation prize. It's a good college and I earned the seat. It took me a while to believe that.

I study Metallurgical and Materials Engineering now. I learn how things are made to carry weight. After the wall, that struck me as funny.

CHAPTER 05 // SIDE 2 // DECISION

HEY YOU

LANE A // CONVEYOR SYMMETRIC

SAME LOOP

Study, test, repeat. The guaranteed four-year loop.

LANE B // SURVEYED UNFINISHED BRANCH

NEW LANE

Build real systems for real people. No safe rails.

TRAJECTORY STATEMENT

Most people my age at college go back to the routine. Study, test, repeat. I get it. It's what got us here.

I didn't want to run the same loop for four more years. So I tried something else: build things for real people and see if they use them.

I don't have a long record yet. I have a habit of starting.

CHAPTER 06 // SIDE 2 // MOMENTUM

THE THIN ICE

ROLE // WEB & TECH TEAM, SPICMACAY NITK ROLE // MARKETING ASSOCIATE, ENGINEER FEST NITK

MOCK TEST ANALYSER

JTestify

React // Vercel

ALGO PAPER TRADING

PaperQuant

Python // Algo

ADAPTIVE DSA ROADMAPS

Transverse

Golang // NextJS

fackNTA — NTA Results Monitor Daemon

pings immediately when NTA updates server

DISPATCH // THE TESTING GROUNDS

I started small and on the side. I joined the web and tech team for SPICMACAY at NITK. I became a marketing associate for ENGINEER, the college fest. I built things on the side.

A mock test analyser called JTestify. An open source paper trading platform, PaperQuant. An adaptive DSA platform, Transverse. And a script, fackNTA, that pings me when NTA posts results. Anyone who has refreshed that page knows why.

None of it came with a guarantee.

CHAPTER 07 // SIDE 3 // HIGH VELOCITY TIMELINE: MAR TO JUN 2026

RUN LIKE HELL

~2M

requests handled by the backend

// DATA CUE: Each red streak represents 1,000 requests.

100k+

aspirants served on AlphaJEE

// DATA CUE: Each square represents one user.

OVERHEAD GANTRY // SHIPPED SYSTEMS

Percentile Predictor

Response Sheet Checker

Mock Test Analyser

RAG Counselling Helper

BACKEND INFRA: Google CloudCloudflareDockerMongoDB

CTO DISPATCH // ALPHAJEE

In March 2026 I co-founded AlphaJEE and became its CTO. We built tools for JEE aspirants. I knew who they were, because I had just been one.

I wrote the backend on Google Cloud and Cloudflare, with Docker and MongoDB. The team shipped a percentile predictor, a response sheet checker, a mock test analyser and a RAG counselling helper. I also handled the community and pitched AlphaJEE to investors and businesses.

By June the backend had handled about 2 million requests for more than 100,000 users.

CHAPTER 08 // SIDE 3 // LESSON

ANOTHER BRICK IN THE WALL (PART 3)

POST-MORTEM // VELOCITY ML

We also built Velocity, an ML-driven, SAT-style adaptive practice platform for JEE. We took it to investors. We couldn't raise money for it.

That stung. Not like the rank did, but it stung. I'm still working out what it taught me. Mostly that building something and getting someone to fund it are two different skills, and I had only practiced one.

The wall was lower this time. I climbed it.

EXHIBIT // VELOCITY ML UNFINISHED

HOLE

Adaptive SAT-style engine for JEE aspirants. Pitch deck delivered, no check closed. Built anyway.

CHAPTER 09 // SIDE 3 // THE FORGE

THE SHOW MUST GO ON

CONSULTANCY & AUTONOMOUS SYSTEMS SURFACE: OXIDE // HOT

MERITVERSE

Autonomous News Agent Pipeline

PHASE 01

RSS Feeds

Continuous stream monitoring across tech & engineering sources

PHASE 02

Poster Engine

Autonomous layout & duotone typography generation

PHASE 03

Instagram Dispatch

Automated direct publication without human touch

Work with Meritverse

FOUNDER DISPATCH // CEO STATEMENT

In June 2026 I started Meritverse and became its CEO. It's a software consultancy. We build web apps and AI agent systems for clients.

One thing we built is an autonomous news agent. It reads RSS feeds, finds a story worth covering, designs a poster for it and posts it to Instagram on its own. Nobody touches it.

I'm a first-year student and a CEO. Some days that sentence sounds ridiculous to me. Then a client writes and I get back to work.

CHAPTER 10 // SIDE 4 // THE EVIDENCE

THE TRIAL

That was the story. Here is the evidence.

CAPABILITIES // TOOLBOX CLUSTERS NO METERS // RAW RIGOR

Agentic AI

LangGraphMCP serversRAG

Web and apps

ReactNextJSNodeJSEdge functionsReact Native

DevOps and data

CI/CDDockerGoogle CloudCloudflarePostgresMongoDBRedis

Languages

JavaScriptPythonGolangRustC++

Runner up, inter-college hackathon

Legisim

An agent-based policy simulator. It splits a country into a few thousand weighted cohorts, applies a policy and traces the effects from the first price change to household budgets.

Explore Case

In Progress

ClankTalk

A language protocol for agents to talk to each other. It drops filler words and vectorizes messages at both ends, to cut token cost and reduce hallucinations.

Agent ProtocolAI CommunicationsIn progressPythonVector DBLLM Protocols

See the case

Transverse

An adaptive DSA platform with personal roadmaps and sandboxed code runs.

EdTechAdaptive LearningSandboxed ExecutionGolangNextJS

See the case

Quickfill-mcp

An MCP server that turns AI chats into live browser pages.

MCP ServerAgentic AIBrowser ToolsTailwindAlpine.jsWASM

See the case

PaperQuant

An open source paper trading platform for algo traders.

Open SourceFinTechAlgorithmic TradingPython

See the case

JTestify

A mock test analysis platform that recommends chapters to study.

AnalyticsEdTechTest PrepReactVercel

See the case

FOR FUN // AUTHENTIC CRAFT

Written without AI help

Some things I write for the fun of it, without AI help. In Rust: a ray tracer, encoders, compressors and parallel SHA-256 brute forcers. Also ESP32 builds, C++ trading scripts and competitive programming.

raytracer.rs Written without AI help

1// Ray-sphere intersection & parallel SHA-256 worker2// Written by hand without AI assistance34pub struct Ray {5    pub origin: Vec3,6    pub dir: Vec3,7}89impl Ray {10    pub fn intersect_sphere(&self, center: Vec3, radius: f32) -> Option<f32> {11        let oc = self.origin - center;12        let a = self.dir.dot(self.dir);13        let b = 2.0 * oc.dot(self.dir);14        let c = oc.dot(oc) - radius * radius;15        let discriminant = b * b - 4.0 * a * c;16        if discriminant > 0.0 {17            Some((-b - discriminant.sqrt()) / (2.0 * a))18        } else {19            None20        }21    }22}