Hello, I'm
Mandeep Singh
Data Scientist
Hello, I'm
Data Scientist
Get To Know More
Rally Vision
Sept-Dec 2024
IISc Bangalore
8.4/10 CGPA
BITS Pilani KK Birla Goa Campus
8.03/10 CGPA
I'm a Data Scientist with a background in Chemical Engineering and
an M.Tech in Computational and Data Science from IISc Bangalore.
My expertise lies in building end to end Machine Learning systems,
from feature engineering and model development to
containerized deployment on cloud infrastructure. I have worked across
computer vision, NLP, RAG, forecasting,
developing solutions that are both technically robust and practical for real world use.
I enjoy applying AI to solve real-world problems and building systems that
are not only accurate, but also scalable, deployable, and production-ready.
What I Work With
Browse My Recent
A full stack platform for organizing and tracking job applications throughout
the hiring process. Users can manage applications, monitor progress across different
stages, and maintain a profile for their job search.
The platform integrates AI powered workflow that generates personalized cover letter
and preparation notes using profile data and job description.
Built with Next.js, TypeScript, MongoDB, FastAPI, and deployed using Vercel, Docker and
AWS EC2.
A domain-specific question answering system built on NVIDIA CUDA Runtime
API documentation using Retrieval Augmented Generation (RAG).
The system combines dense retrieval using OpenAI embeddings and ChromaDB
with BM25 sparse retrieval, fused through Reciprocal Rank Fusion (RRF).
Retrieval quality is further improved through multi-query expansion and
Cohere reranking, enabling accurate answers grounded in the official CUDA documentation.
A machine learning system for forecasting daily retail sales
using the Rossmann Store Sales dataset.
The project includes exploratory data analysis, feature engineering,
and model development using XGBoost with Optuna-based hyperparameter
optimization. Historical sales patterns, promotions, holidays, competition,
and store metadata are used to generate forecasts, which are
visualized through an interactive Streamlit dashboard.
An image classification system for fungi image classification
using deep learning and transfer learning techniques.
The model is built on EfficientNetV2 with selective fine-tuning
of deeper layers and a custom classification head. Data augmentation
and class-weighted loss were used to improve generalization on an
imbalanced dataset, achieving 90.7% classification accuracy.
The final model was exported to ONNX for efficient deployment and inference.
A sentiment analysis system trained on the IMDb movie reviews
dataset to classify reviews as positive or negative.
The project compares transformer based and classical machine learning approaches,
including DistilBERT and TF-IDF-based models. DistilBERT achieved 90.86% accuracy,
while a TF-IDF + Logistic Regression model (88.03% accuracy) was selected for deployment to
provide faster and more cost-efficient inference.
An audio classification system that predicts music genres from
audio clips using both handcrafted features and deep audio embeddings.
The project explores traditional audio features extracted with Librosa alongside
pretrained YAMNet embeddings. Multiple machine learning models were evaluated,
with YAMNet embeddings achieving 85% accuracy. For deployment,
a lightweight Librosa + Logistic Regression pipeline (78% accuracy) was chosen
to balance performance with computational efficiency.
Get in Touch
