FinRAG: Retrieval-Augmented Generation for Financial QA
Built a financial-domain RAG system that answers investment, risk-analysis, and forecasting queries with improved factuality and reduced hallucinations.
Hi, I'm Suraj Thapa
I develop robust estimation pipelines and explore machine learning/computer vision.
About Me
I am a Master's student in Data Science at the University of New Haven, deeply passionate about building intelligent, robust systems that can solve complex real-world problems. My academic and professional journey is driven by a fascination with transforming raw data into actionable, high-impact intelligence.
Rather than just training models, I enjoy the entire end-to-end process of data science—from architecting efficient data pipelines and engineering software, to deploying scalable machine learning solutions. I thrive in dynamic environments where rigorous theoretical mathematics meets practical, hands-on software engineering.
Beyond my academic coursework, I am a strong advocate for the open-source community. I value continuous learning, cross-disciplinary collaboration, and breaking down complex topics through detailed technical writing and tutorials.
Research Interests
I explore robust estimation and computer vision, with a focus on accuracy, determinism, and reproducibility.
Robust Estimation
RANSAC and deterministic variants to effectively filter extreme outliers.
Computer Vision
Homography, object tracking, and instance segmentation techniques.
Information Fusion
Fusing multi-modal sensory inputs like LiDAR and RGB arrays for SLAM.
Machine & Deep Learning
Training specialized neural networks, optimization algorithms, and NLP applications.
Environmental Science
Applying data-driven and geospatial techniques directly to ecological conservation and analysis.
Projects
Selected academic and applied work across vision, ML, and systems.
Instance Segmentation Using Mask R-CNN
Implemented a three-class instance segmentation pipeline on a custom dataset using Detectron2/Mask R-CNN. Trained & fine-tuned models analyzing mAP, IoU.
Tango Puzzle Solver with AC-3 and A* Search
End-to-end solver for the Tango logic puzzle with interactive visualization via Pygame. Extended with Q-learning experiments.
EDA and DE Pipeline
Exploratory Data Analysis and Data Engineering pipeline ensuring robust transformation, cleaning, and preparation of raw datasets for scalable machine learning.
ORB-SLAM3 Experiments
Explorations and integrations using the ORB-SLAM3 framework for visual, visual-inertial, and multi-map SLAM operations across diverse environments.
Curriculum Vitae
Blog
Insights and tutorials on deep learning, machine learning algorithms, and optimization.
Visual Guide: 7 Activation Functions Explained
A comprehensive breakdown of ReLU, Sigmoid, Tanh, and other critical neural network activation functions with generated mathematical plots and use-cases.
Gradient Descent & Its Variants (BGD vs SGD)
Understanding the calculus behind model optimization and why mini-batches strike the perfect balance between speed and stability.
Stop Relying on Accuracy: Real Evaluation Metrics
Why the Confusion Matrix, Precision, Recall, and F1-Scores are far more important than raw Accuracy—especially with imbalanced datasets.
Principal Component Analysis (PCA) Intuition
Defeating the Curse of Dimensionality by using variance-maximizing orthogonal transformations to compress massive datasets.
Law of Large Numbers vs Central Limit Theorem
Unpacking two of the most commonly confused fundamental theorems in classical probability and how they power data science inference.
Get in Touch
I am always open to discussing research collaborations, internships, or data science opportunities. Send me a message!