Ankit Dey

Hey! I'm Ankit, a Computer Engineering student who builds AI systems and web apps that people actually use. I care about making things fast, reliable, and practical, not just impressive on paper.

Book a call

GitHub↗LinkedIn↗X↗

Location

Kolkata, India

Building

Cool stuff. Always.

Focus

AI Systems & Web Scale

Subject_Profile // 0x4A1E

about

A backend dev and AI/ML engineer who loves building stuff that actually works at scale. I've shipped Discord bots like that hit 200k+ users, and built that use computer vision and GenAI to make sense of messy, real-world data. I spend most of my time making things fast, reliable, and actually useful, not just technically impressive.

Current Focus

Building with LLMs & RAG

Philosophy

Make it work, then make it fast

Status

Down to collaborate

Expertise

Web Systems, AI Agents

education

The formal education path. Still learning, always will be.

workspace_premium

St Joseph's School

Secondary Education

COMPLETED

history_edu

Oriental Public School

Higher Secondary Education

COMPLETED

school

Narula Institute of Technology

Bachelor of Technology in Computer Science

2024 - 2028

work experience

Where I've been building things that matter. Each one taught me something new.

achievements

Highlights

Some highlights from along the way. Each one paid off.

Recognition

Member of Swiggy Builders Club

Exclusive Club

Got into Swiggy's exclusive Builders Club for building an OAuth-secured, MCP-powered Swiggy integration with Discord. Pretty cool to be recognized for something that was just a fun experiment.

2026

Hackathon

Smart India Hackathon Finalist

National Finalist

Made it to the national finals with a real-time attention monitoring system using computer vision. Basically taught a camera to know when you're actually paying attention.

2025

Platform Recognition

Verified Discord Bot

200k+ Users

Got Sync Music verified by Discord after it hit 200k+ users. Started as a side project, turned into something people actually paid for. Still kind of surreal.

2024

Hackathon

Hack-O-NiT Finalist 2026

West Bengal

Made it to the finals at Hack-O-NiT, competing against some really smart people from across West Bengal. Great experience, learned a ton.

2026

projects

Count: 04

A selection of projects spanning web systems, AI pipelines, and scalable infrastructure.

200k+ active users

Sync Music

Verified Discord music bot for 200k+ users.

•Live

Scalable Discord.js v14 + TypeScript architecture with a pluggable command/event layer, MongoDB persistence, Lavalink audio streaming, and premium/vote-gated controls.

Discord.jsTypeScriptMongoDBLavalinkNode.js

INDEX

RAG pipeline

TubeInsight AI

Comment intelligence and insight discovery.

•Repo

End-to-end ingestion, preprocessing, embeddings, semantic indexing, KMeans theme discovery, and a RAG chat layer for fast feedback exploration.

EmbeddingsKMeansRAGNLPVectorDBs

94.3% accuracy

landmarks // ear_ratio

E.A.R and Face Landmark Detection

Real-time attention monitoring with CV.

•Live

Python, MediaPipe, and OpenCV system using 468-point facial landmark tracking and Eye Aspect Ratio blink scoring for attention detection.

Computer VisionMediaPipeOpenCVPython

builders club

Swiggy MCP Discord Bot

Builders Club prototype for Swiggy inside Discord.

•Build

Discord.js v14 and TypeScript bot that connects Swiggy Instamart through MCP with OAuth 2.1 PKCE login, secure token lifecycle handling, and slash-command auth flows.

Discord.jsTypeScriptMCPOAuth 2.1Express

200k+ active users

Sync Music

Verified Discord music bot for 200k+ users.

•Live

Scalable Discord.js v14 + TypeScript architecture with a pluggable command/event layer, MongoDB persistence, Lavalink audio streaming, and premium/vote-gated controls.

Discord.jsTypeScriptMongoDBLavalinkNode.js

INDEX

RAG pipeline

TubeInsight AI

Comment intelligence and insight discovery.

•Repo

End-to-end ingestion, preprocessing, embeddings, semantic indexing, KMeans theme discovery, and a RAG chat layer for fast feedback exploration.

EmbeddingsKMeansRAGNLPVectorDBs

94.3% accuracy

landmarks // ear_ratio

E.A.R and Face Landmark Detection

Real-time attention monitoring with CV.

•Live

Python, MediaPipe, and OpenCV system using 468-point facial landmark tracking and Eye Aspect Ratio blink scoring for attention detection.

Computer VisionMediaPipeOpenCVPython

builders club

Swiggy MCP Discord Bot

Builders Club prototype for Swiggy inside Discord.

•Build

Discord.js v14 and TypeScript bot that connects Swiggy Instamart through MCP with OAuth 2.1 PKCE login, secure token lifecycle handling, and slash-command auth flows.

Discord.jsTypeScriptMCPOAuth 2.1Express

blogs

#Security #OAuth

How Do You Prove You're Actually You? PKCE and SHA-256 Explained Simply

A practical deep dive into OAuth 2.1 PKCE flow and SHA-256 hashing from building a production Discord bot with Swiggy integration...

#AI #Evolution

Evolution Is Back: A New Way to Fine-Tune LLMs

Rediscovering Evolution Strategies as a scalable alternative to reinforcement learning for fine-tuning billion-parameter LLMs, with massive GPU efficiency gains through EGGROLL...