Hello, I'm
Bibek
01
About Me
Data Science
Student
Data Science major at Southeastern, with a habit of picking up whatever the project needs. Usually ends up being React and TypeScript for UIs, ASP.NET for APIs, and Python for anything involving data. I build things end-to-end — not because it's always the efficient choice, but because I want to understand the whole thing. Right now I'm deep in computer vision and on-device ML, which turned out to be a much larger rabbit hole than expected.
02
Skills
Area of
Expertise
Python
Where most of my actual work happens. ETL pipelines, ML experiments, the occasional one-off script that I only meant to run once.
React
TypeScript, hooks, context. I stopped fighting the re-render cycle a while back and started working with it instead.
JavaScript & TypeScript
I defaulted to plain JavaScript until TypeScript made me feel bad about it. Now I use TypeScript and things make more sense.
ASP.NET / C#
Where my APIs live. C# makes you be explicit about types in a way I've come to appreciate, even when it adds an extra line or two.
Full-Stack & APIs
Connecting all of the above. The part where frontend and backend disagree about who's responsible for the bug.
Also familiar with
React Native
Node.js
Flask
NumPy
Matplotlib
Pandas
TensorFlow
OpenCV
Scikit-Learn
Scikit-Image
Keras
Java
C++
JavaFX
HTML5
CSS3
MySQL
Git
03
Projects
Things I've
Built
02
WeSchedule
Full-Stack Web App
A group calendar where you actually control who can edit what. Built because shared scheduling tools always feel like they were designed for someone else's workflow. Ended up implementing a full permissions system, which taught me more about auth than any tutorial had.
03
Flight Crash Analysis
Data Science / Python CLI
A Python CLI that breaks down fatal aviation crash data from 2019–2025 — ETL pipeline, statistical summaries, charts, conditional probabilities. Started as a class project. The patterns in the actual data turned out to be more interesting than I expected.
04
Project A+
Mobile App · Computer Vision · On-Device AI
Started as a hackathon idea: scan your notes on your phone and ask it questions. Grew into an OpenCV preprocessing pipeline (Canny edge detection, Sauvola thresholding, perspective correction), on-device AI for offline use (Gemma 4:E2B via GGML), and a routing layer that falls back to Gemini 2.5 Pro when there's a connection. Generates Q&A, quizzes, and flashcards. More moving parts than planned — but it works without Wi-Fi, which was the whole point.
05
ECG Biometric Auth
Deep Learning · Biometrics · Signal Processing
Turns out your heart's electrical signature is unique enough to use as a password — and harder to steal than a fingerprint. Built a 1D ResNet encoder that compresses individual heartbeats into 128-dimensional embeddings using Triplet Loss, so same-person signals cluster and impostors don't. Trained on 21k PTB-XL ECG recordings. Reached 0.972 AUC and 9.33% EER — good enough to be interesting, imperfect enough to be honest about. Open source under GPLv3.