ML student · AI safety enthusiast · Building things to understand how models really work
I'm learning machine learning by actually building with it. Not just running code, but trying to understand what's happening inside the models.
Lately I've been drawn to AI safety: how LLMs respond to prompts, where hallucinations come from, and how user intent gets interpreted (or misinterpreted). I'm experimenting with systems that classify inputs as safe, unsafe, or suspicious.
- Building ML projects from scratch to understand models at a deeper level
- Exploring prompt behaviour, injection risks, and hallucination patterns in LLMs
- Experimenting with input classifiers (safe / unsafe / suspicious)
- Improving at debugging models, not just deploying them
Core
Python · SQL
Data & ML
NumPy · Pandas · Matplotlib · Scikit-learn
Focus areas
LLMs / NLP · AI Safety · Prompt Engineering · Computer Vision
Prompt-Safety-Classifier[https://github.com/leeeshart/Prompt-Safety-Classifier]
A system that classifies LLM inputs as safe, unsafe, or suspicious. Explores prompt injection patterns and hallucination triggers in large language models.
Python · LLMs · NLP · Transformers
Ecommerce-sales-data-analysis-[https://github.com/leeeshart/Ecommerce-sales-data-analysis-]
Exploratory data analysis with visualisations — patterns, surprises, and what the data actually reveals beyond the summary stats.
Pandas · Matplotlib · SQL
IMS Ghaziabad — University Courses Campus
"I'm learning by building, experimenting, and documenting what works — and what doesn't."