Projects | Thibaut Kivuvu

🕸️ CS50W Projects

F1 Analyst (Capstone)

Full-stack Django web app for exploring Formula 1 data. Integrates the Ergast/Jolpica F1 API to display schedules, results, and stats with clean, data-driven pages and performance-minded caching.

Why it’s strong: Django architecture, external API integration, reusable services, and UI built for real users.

GitHub

Django views + templates, structured routing, and reusable service layer for API calls.
API-driven pages for drivers, constructors, race calendar, and results with sensible caching.
Responsive layout with Bootstrap and dynamic navigation components.

Network (Social Network)

Django social network app featuring user profiles, following/unfollowing, post creation, likes, and pagination. Uses JavaScript (fetch/AJAX) for dynamic interactions.

Why it’s strong: auth + relational data modeling + asynchronous UI updates.

GitHub

User authentication, profile pages, and follow graph relationships.
Likes and edit-post features driven by fetch requests without full page reloads.
Pagination and clean templates for readability and scalability.

Mail (SPA Email Client)

Single-page email client built with JavaScript. Loads Inbox, Sent, and Archive views asynchronously and supports reading, replying, and archiving messages through API calls.

Why it’s strong: front-end architecture, async fetch workflows, and state-driven UI.

GitHub

SPA navigation with dynamic rendering for compose, mailbox lists, and detail view.
Fetch requests for read/unread state and archive toggles.
Reply flow that pre-fills and formats message context.

Commerce (Auctions)

eBay-style auction platform with Django models for listings, bids, comments, and categories. Includes watchlists, bidding logic, and winner determination at close.

Why it’s strong: database-driven features, forms, validation, and business rules.

GitHub

Django ORM modeling for listings, bids, comments, categories, and watchlists.
Authorization rules for bidding/closing and watchlist UX.
Clean templates with reusable components and validation feedback.

🧠 CS50AI Projects

Traffic

Built a convolutional neural network in TensorFlow/Keras to classify traffic sign images across multiple categories. Focused on image preprocessing, model design, training, and evaluation.

Why it’s strong: practical machine learning workflow, computer vision, and model tuning.

GitHub

Preprocessed image data into a format usable for CNN training.
Built and trained a TensorFlow/Keras model with convolution, pooling, dense, and dropout layers.
Strengthened understanding of overfitting, architecture choices, and model evaluation.

Attention

Worked with transformer-based language modeling concepts by analyzing and visualizing self-attention patterns in masked language processing tasks.

Why it’s strong: modern NLP exposure, transformer concepts, and attention visualization.

GitHub

Used a transformer model to inspect masked token prediction behavior.
Generated visual attention diagrams for layers and heads.
Built intuition for how modern NLP models represent relationships between words.

Nim

Implemented a reinforcement learning agent using Q-learning to learn how to play Nim through self-play. Focused on state/action values, exploration, and iterative improvement.

Why it’s strong: reinforcement learning fundamentals and algorithmic problem solving.

GitHub

Implemented Q-value updates, best-future-reward logic, and epsilon-greedy action selection.
Trained the agent through repeated self-play games.
Demonstrated how an agent can improve performance from experience rather than hard-coded rules.

Shopping

Built a machine learning classifier to predict online shopping revenue outcomes from session data using scikit-learn, then evaluated model performance with sensitivity and specificity.

Why it’s strong: applied classification, feature handling, and evaluation metrics.

GitHub

Prepared structured tabular data for a supervised learning model.
Used K-nearest neighbors with scikit-learn for binary classification.
Evaluated results using performance metrics beyond simple accuracy.

PageRank

Implemented the PageRank algorithm using both sampling and iterative approaches to estimate the importance of pages in a linked corpus.

Why it’s strong: probability-based reasoning, graph algorithms, and search foundations.

GitHub

Modeled transition probabilities between linked pages.
Computed rankings through both random sampling and iterative convergence.
Strengthened understanding of ranking systems and probabilistic search methods.

Crossword

Built a crossword generator using constraint satisfaction techniques such as node consistency, arc consistency, heuristics, and backtracking search.

Why it’s strong: optimization, constraints, and structured search strategies.

GitHub

Applied node and arc consistency to reduce the search space.
Used heuristics such as MRV, degree, and least-constraining-value ordering.
Implemented backtracking search to construct valid puzzle solutions.

🐍 CS50P Projects

F1 History CLI

Python command-line app that pulls Formula 1 race and qualifying results via the Jolpica/Ergast F1 API. Supports driver and constructor standings and includes a pytest test suite.

Why it’s strong: real-world API integration, structured CLI UX, and automated testing.

GitHub

Parses JSON payloads with graceful error handling.
Reusable API wrapper and CLI argument parsing.
pytest coverage for endpoints, parsing, and formatting.

Jar (OOP Demo)

Python class that models a jar with capacity, add/remove methods, and current fill state. A concise example of object-oriented design, validation, and testable program structure.

Why it’s strong: clean OOP design, state validation, and unit-test-friendly logic.

GitHub

Constructor enforces capacity rules.
Guard clauses prevent negative fill states and overflow.
Good teaching sample for class design and testing.

🌐 CS50x Projects

NBA Stats Hub

Flask web app that retrieves NBA player statistics from a public API, aggregates data, and displays visual charts, search filters, and team logos.

Why it’s strong: full-stack app, API + database + templating, and interactive UI.

GitHub

Player query assistant with AI-ready architecture.
Light/dark theme toggle and responsive layout.
Caching and combined stat logic for multi-team seasons.

Finance

Simulated stock trading platform with register/login, quote lookup, buy/sell, transaction history, and portfolio value, built with Flask and SQLite.

Why it’s strong: authentication, session management, database CRUD, and input validation.

GitHub

Password hashing and session management.
Real-time stock lookups via API.
Transaction ledger and portfolio aggregation.

Sports Homepage & Quiz

Responsive multi-page static site highlighting favorite NBA and football players and teams, plus an interactive sports quiz built with vanilla JavaScript.

Why it’s strong: responsive layout, front-end theming, and client-side interactivity.

GitHub

Multi-page navigation with dropdown menus.
Interactive quiz with JavaScript click handlers and instant feedback.
Bootstrap styling with custom CSS overrides.
Accessible headings and semantic markup.

Fiftyville SQL Mystery

Investigative SQL project querying a relational database of phone logs, flights, and ATM use to solve a fictional crime through filtering, joins, and inference.

Why it’s strong: practical SQL reasoning, multi-table joins, and data sleuthing.

GitHub

Joined evidence across multiple related tables.
Used stepwise elimination querying to narrow suspects.
Summarized findings tied to the narrative solution.