Hello, I'm
Chay Harjit.

AI Builder & Ethical Hacker

A little context

About
Me

I'm Chay Harjit, a Grade 9 student at Hayagrivas International School, exploring Artificial Intelligence through Raam TechLink. I enjoy building useful AI projects and learning how cybersecurity can be made more thoughtful and effective.

01 Student builder 02 AI explorer 03 Security minded

Projects

Education · AI platform

LearnSub

AI-powered personalized learning for every student, in every language.

View

Core problem

When students miss school, they can lose foundational concepts and fall further behind, especially without tutors or when learning in a non-native language.

Key features

Upload PDFs, videos, or notes for automatic concept extraction
Adaptive quizzes with instant feedback
Study across 100+ languages
Real-time AI chat tutor and progress dashboards

Unique value

Purpose-built educational AI with true personalization, language-first design, and one web-based platform that scales from individual learners to institutions.

Machine learning · Student wellbeing

Conduct AI

ML-powered student behavior assessment for earlier, more informed support.

The problem

Social disengagement, declining academic behavior, and sudden changes are often missed by traditional monitoring, delaying intervention.

Solution overview

Behavioral risk prediction from 15 key features
XGBoost classification for risk levels
Interactive assessment interface for real-time evaluation

Engineering highlights

Balanced data, engineered features, and optimized training for stronger performance. Robust session state keeps the assessment experience smooth and reliable.

Live completion trackingLow / Medium / HighVisual decision support

Agriculture · Responsible AI

Agrochemicals Predictor

AI-powered fertilizer recommendations for safer, smarter, and more sustainable agriculture.

View

The challenge

Improper fertilizer and pesticide use can create health risks, pollute soil and water, reduce fertility, and affect crop yields. Farmers often lack precise, data-driven guidance.

Objective

Predict safe and optimal fertilizer dosages from crop type, growth stage, soil properties, and environmental conditions.

Recommendation inputs

Crop type and growth stage
Soil type, pH, moisture, and nutrient levels
Temperature, humidity, and rainfall

Education · AI Assessment

Xaminr

Introducing Xaminr - an AI-powered exam correction tool.

View

How it works

Teacher uploads the Answer Key
Teacher uploads the Student's Paper
AI reads, evaluates, and corrects instantly

Results in seconds

Marks & Percentage
Grade (A/B/C/D/F)
Pass / Fail

Detailed Feedback

Question-by-question feedback
Strengths & areas to improve

What I work with

Skills

Curiosity across intelligent systems, security, and the tools that make ideas real.

8 disciplines3 practice areas

Intelligence

Artificial IntelligenceMachine LearningAI Agents

Security

Ethical HackingCyber Security

Tools & systems

PythonLinuxKali Linux

Creative

Designer

Task

Security · Ethical Hacking

Credential Harvesting

Demonstrating Credential Harvesting Techniques in Kali Linux. A local proof-of-concept run in my lab environment.

View

Execution & Setup

Launched zphisher.sh terminal tool in Kali Linux.
Selected the Google target option and hosted a local login page.
Set up hosting and port forwarding.

Testing & Capture

Opened local link to load fake Google Sign-In.
Successfully captured login info and victim IP.

AI · Fine-tuning

Phi-4-Reasoning 14B

Used Unsloth AI + QLoRA (4-bit) to fine-tune a 14B reasoning model on a custom JSONL dataset.

View

Tech stack

Model: Unsloth AI/phi-4-reasoning (4-bit quantized)
Fine-tuning: LoRA rank 32 across all attention + MLP layers
Trainer: Hugging Face TRL SFTTrainer
Dataset: Custom JSONL in chat-format

AI · LangChain

URL RAG Agent

A URL RAG Agent using Gemini & LangChain. Paste any webpage URL and instantly chat with its content using AI. It fetches, processes, and lets you ask questions naturally.

View

Useful for

Research & documentation
Competitive analysis
Learning new topics
Summarizing long articles

AI · RAG

End-to-end RAG Pipeline

An end-to-end RAG pipeline that answers user queries by dynamically searching the web, extracting content, and generating AI-powered responses.

View

How it works

Takes a user query as input
Searches the web using DuckDuckGo
Loads and extracts content from top URLs
Splits documents into chunks for context handling
Converts text into embeddings and stores in FAISS
Retrieves relevant chunks via similarity search
Sends context + question to Gemini for final answer

Key Features

Real-time web knowledge retrieval
Lightweight and fast embedding model (MiniLM)
Modular LangChain pipeline design
Fully automated RAG workflow (no manual prep)

Security · Ethical Hacking

Metasploit Framework Lab

Completed a hands-on cybersecurity lab using Kali Linux and the Metasploit framework to understand how attackers exploit system vulnerabilities.

View

Simulated Attack

In a controlled and ethical environment, I simulated an attack on a Windows machine by creating a custom payload to demonstrate how malicious files can run silently in the background without obvious signs to the user. This allowed me to observe how system information could potentially be accessed if security measures are weak.

Defense & Insights

I also explored the concept of Remote Access Tools (RATs) from a defensive perspective, gaining insight into how such techniques are used and how they can be detected and prevented. This experience strengthened my understanding of both offensive and defensive security practices.

AI · RAG

Custom AI/ML Glossary RAG

I built my own Retrieval-Augmented Generation (RAG) System with Gemini + LangChain. This approach significantly reduces hallucinations and answers questions based strictly on custom data.

View

What it does

Creates a custom AI/ML glossary (data.txt)
Splits documents into chunks
Generates embeddings & stores in a vector DB
Retrieves the most relevant context
Generates grounded answers using an LLM

Key Concepts Used

Embeddings
Vector Search
Prompt Engineering
Context Injection
LLM Orchestration

Cybersecurity · Practice

Hands-On Wireless Security Assessment

I completed a practical lab focused on wireless network security testing using industry-standard tools in a controlled lab environment.

View

Key Activities

Configured a wireless adapter in monitor mode
Captured and analyzed 802.11 traffic
Observed WPA2 authentication handshakes
Performed controlled deauthentication testing in a lab setup
Conducted password strength testing using a wordlist (for security evaluation purposes)
Analyzed packet captures using Wireshark

AI · Computer Vision

Hamster Object Detection Model

I built a computer vision project focused on building an Object Detection model to accurately identify hamsters using Roboflow. This project involved designing and implementing a complete end-to-end workflow — from dataset preparation to model evaluation.

View