_01. Portfolio — Calgary, AB

HappyPrajapati

AI/ML Engineer & Full-Stack Developer. Building deep learning pipelines and production web applications — from CNN benchmarking to full-stack systems.

GitHub ↗LinkedIn ↗

_02. About

I'm a Computer Science student at the University of Calgary, splitting my time between deep learning research and production software engineering. I care about pipelines that hold up under real data, and interfaces that get out of the user's way.

Education

Sept 2024 – Present

Bachelor of Science, Computer Science

University of Calgary, Alberta

Minor in Mathematics · Science Co-op Program

Education

Sept 2023 – Apr 2024

Bachelor of Science, Computer Science

University of Ottawa, Ontario

First year — transferred to University of Calgary

96.8%

Model accuracy, DANSA Lab CV pipeline

6

Deep learning architectures benchmarked

50+

Automated test programs, compiler suite

5

Engineers led, OMG platform team

_03. Stack

Tools I reach for

Languages & Web

12
PythonSQLJavaCJavaScriptTypeScriptHTML/CSSReactJSNext.jsTailwind CSSGSAPFramer Motion

Machine Learning & AI

08
TensorFlowPyTorchKerasCNNComputer VisionDeep LearningPandasNumPy

Backend & Cloud

05
RESTful APIsFastAPIPostgreSQLMySQLAWS Lambda

Tools & DevOps

06
GitGitHubGitLabLinuxCI/CD PipelineVercel

Testing & QA

04
PytestIntegration TestingUnit TestingSDLC

AI Tools & Development

06
GitHub CopilotChatGPTClaudeLLMsPrompt EngineeringGenerative AI

_04. Experience

Where I've worked

Jan 2026 – Present

Calgary, AB

Undergraduate Researcher

DANSA Lab, University of Calgary

  • Achieved 96.81% accuracy building and deploying a production-quality Python deep learning pipeline to ingest, process, and classify large-scale unstructured video datasets for real-time anomaly detection.
  • Benchmarked six deep learning architectures (EfficientNetB0, VGG16, DenseNet121, ResNet50, MobileNetV2, EfficientNetB2L) for video frame image classification and accuracy configurations to identify the optimal model.
  • Applied an evaluation-driven approach using ROC/PR curves and validation metrics to define benchmarks, identify failure modes, and improve AI pipeline performance across experimental configurations.

_05. Selected Projects

Applied engineering

Case studies from coursework and team initiatives — architecture, ownership, and measurable outcomes for each.

_01

This Portfolio

Aug 2026

Personal Project · Solo project

This site — a Next.js portfolio with a single content-driven data file, scroll-triggered animation, and a from-scratch reload sequence, built to be easy to extend as new projects and research come in.

Architecture

  • Next.js App Router site with all content sourced from one data file, so sections update without touching component code.
  • GSAP ScrollTrigger reveals, a custom cursor, and a canvas-based starfield background, each gated behind prefers-reduced-motion checks.
  • Horizontal scroll-snap project carousel built to scale as more case studies get added.

Metrics

Next.js 15 App Router + TypeScript
0 ESLint / build errors at ship time
Next.jsTypeScriptTailwind CSSGSAPFramer Motion
Live ↗GitHub ↗
_02

Python-to-MIPS Compiler

Jan 2026 – Apr 2026

CPSC 411 · Solo project

A from-scratch compiler translating a Python-like source language down to MIPS assembly, covering the full pipeline from lexical analysis through low-level code generation.

Architecture

  • Lexical analysis and parsing to build an AST from source input.
  • Semantic analysis and type checking passes to catch errors before code generation.
  • Lowering from the AST to MIPS assembly output.

Metrics

50+ programs validated via automated Pytest suite
MIPS target instruction set architecture
PythonCompiler DesignMIPS AssemblyPytest
GitLab ↗
_03

LibraTech — Library Management System

Nov 2025 – Dec 2025

Team Project · Small team

A library management system built with vanilla HTML, CSS, and JavaScript, giving readers a searchable catalog with checkout tracking and giving librarians an admin dashboard to manage inventory and due dates.

Architecture

  • Catalog browsing and search interface for readers to find and check out titles.
  • Checkout tracking to keep book availability and due dates current.
  • Admin dashboard for librarians to manage inventory and monitor due dates.

Metrics

2 views reader catalog + librarian admin dashboard
HTMLCSSJavaScript
Live ↗GitLab ↗
_04

OMG — Online Multiplayer Game Platform

March 2025 – Apr 2025

Team Initiative · Led a 5-person unit within a 25-student initiative

Real-time Java components for multiplayer game-state management, matchmaking, and live leaderboard processing, built as part of a larger cross-team initiative spanning three shared games.

Architecture

  • Java software components built with Maven for game-state management and matchmaking.
  • Live leaderboard processing shared across three games developed by the wider initiative.
  • Cross-team requirements gathering to keep shared systems compatible across game teams.

Metrics

5 engineers led directly
3 shared games integrated
25 students across the initiative
JavaMavenReal-time SystemsIntegration Testing
GitLab ↗
_05

TrustBuy — Online Marketplace

Jan 2025 – Apr 2025

Full-Stack Application · Team project

A full-stack marketplace application integrating a Django REST API backend, MySQL persistence, and a ReactJS frontend into scalable, production-ready user workflows.

Architecture

  • Django REST APIs for backend logic, connected to a MySQL data layer.
  • Functional specifications translated into reusable React components for maintainable frontend features.
  • Refactored REST APIs and database schema to improve data transformation, scalability, and reliability.

Metrics

Full-stack backend, frontend & API ownership
Django REST FrameworkMySQLReactJS
GitHub ↗

_06. Research

Academic work

Research conducted through DANSA Lab, kept separate from applied coursework projects above.

Manuscript in preparation — under publicationJan 2026 – Present

Suspicious Activity Identification from Video Surveillance Data using Fine-tuned Convolutional Neural Networks and EfficientNetB0

DANSA Lab, University of Calgary

Research building a production-quality deep learning pipeline that ingests, processes, and classifies large-scale unstructured video datasets for real-time anomaly detection, benchmarking six CNN architectures and using ROC/PR analysis to define evaluation criteria.

96.8%

peak classification accuracy

6

architectures benchmarked: EfficientNetB0, VGG16, DenseNet121, ResNet50, MobileNetV2, EfficientNetB2L

ROC / PR

evaluation methodology

Preprint link coming on publication