Samuel CoqRésumé (PDF, opens in new tab)

SAMUEL COQ / SOFTWARE & APPLIED AI

Less busywork.
More possibility.

I’m Samuel, a Computer Science student building AI and automation tools that make everyday tasks easier.

UMass Lowell · Graduating December 2027 · Seeking software and AI internships

Explore my work GitHub (opens in new tab)
BASED IN LOWELL, MAPYTHON C / C++ APIS MACHINE LEARNINGSCROLL TO EXPLORE

01 / Selected work

Real problems. Working ideas.

What I built, how it works,
and what I’m learning.

03 projects
Outcome design
SiS registration botInteractive design preview
Check seats
Read result
Notify
Enrolled

The portal confirms enrollment. This course can stop being monitored.

Choose a sample portal outcome

Illustration only. Scheduling and notification-loop integration are still in development.

01 / AutomationStarted Jul 2026

Active development

SiS Course Registration Bot

A seat without the endless refresh.

A Python and Playwright prototype that checks course seats and, when it finds an open one, attempts enrollment and reads the portal’s actual response.

  • Python
  • Playwright
  • n8n
4explicit enrollment outcomes
View code (opens in new tab)
Explore the case study: SiS Course Registration Bot

The problem

Checking seats gets repetitive.

High-demand courses can fill quickly. I wanted a tool that could check requested courses and distinguish a successful enrollment from a rejection or an automation failure.

What I built

Let the portal explain the result.

The prototype handles a signed-in browser session, checks seat availability, and returns enrolled, rejected, already enrolled, or error. A separate n8n-to-Gmail notification pipeline has been built and verified end to end.

What comes next

Connect the pieces reliably.

The repository still lists scheduled polling, notification integration into the live loop, and permission-code handling as unfinished. The current prototype requires supervision; the workflow preview illustrates the outcome design.

02 / Machine learningAug 2025

Completed study

Airbnb Instant Booking Predictor

From listing data to model comparisons.

A classification study across 28,022 Airbnb listings, comparing Logistic Regression and Decision Tree models with recorded notebook results.

  • Python
  • Pandas
  • Scikit-learn
  • Matplotlib
28,022listings in the original dataset
View code (opens in new tab)Open notebook (opens in new tab)
Explore the case study: Airbnb Instant Booking Predictor

The question

Which listings allow instant booking?

I used host, listing, and review features to classify the instant_bookable label. The dataset contains 20,382 non-instant-bookable listings and 7,640 instant-bookable listings, so accuracy needs class-balance context.

The approach

Prepare, train, compare.

The notebook includes median imputation, one-hot encoding, removal of constant features, a 75/25 train/test split, and five-fold grid search. I compared default and tuned Logistic Regression and Decision Tree models.

The result

A result with room to improve.

The notebook records 75.46% accuracy for the tuned tree with five selected features. Preprocessing and feature selection were fitted on the full dataset, so this is not a leakage-free estimate. The next step is to fit those stages inside training folds and reevaluate on untouched test data.

From the original notebook

The result, with its context.

Recorded outputs from the public project notebook. These models were not rerun for this website.

28,022listings
75 / 25train / test split
72.7%non-instant-bookable share of the full dataset
Accuracy recorded in the notebook’s outputs
ModelRecorded accuracy
Default Logistic Regression72.85%
Default Decision Tree67.21%
Tuned Decision Tree75.34%
Tuned Decision Tree + 5 selected features75.46%

Evaluation limitation: preprocessing and feature selection were fitted on the full dataset. These scores need a fresh evaluation with preprocessing and selection fitted only inside training folds. The 72.7% class share above is dataset context, not a measured test-set baseline.

Original notebook ROC curves comparing default, tuned, and feature-selected Logistic Regression models.
Logistic Regression comparisons · original notebook output
Original notebook ROC curves comparing default, tuned, and feature-selected Decision Tree models.
Decision Tree comparisons · original notebook output
03 / AutomationIn development

In progress

Canvas Homework Tracker

A study plan that can fit real life.

An AI agent in progress to turn Canvas assignment deadlines into prioritized study plans and Google Calendar time blocks.

  • Python
  • Claude API
  • Google Calendar API
Explore the case study: Canvas Homework Tracker

The problem

A deadline isn’t a study plan.

Knowing when an assignment is due doesn’t tell you when to work on it. I’m developing a workflow around the Canvas agenda feed, deadlines, and estimated effort.

In development

From assignment to time block.

The planned agent loop uses the Claude API to draft study plans and the Google Calendar API to find open blocks and schedule study sessions. The integration is still being developed.

Planned next

Make space for changing plans.

Automatic replanning after missed sessions and reminder notifications remain planned features. This project is presented as work in progress, rather than a completed product.

Thoughtful tools start with a problem worth solving.

02 / THE PERSON BEHIND THE CODE

Curious by nature.
Practical by choice.

Hi, I’m Samuel, a Computer Science student at UMass Lowell with a Data Science concentration and a Mathematics minor, graduating December 2027.

My interests are agentic systems, automation, machine learning, and software development. I work in English, French, and Haitian Creole, which is useful more often than I expected.

Read my full résumé (PDF)
FOUNDATIONS

UMass Lowell

B.S. Computer Science · Data Science
3.69 GPA · Expected Dec 2027

APPLIED LEARNING

Break Through Tech AI at MIT

AI fellowship · May 2025 – Apr 2026
Model development, evaluation & responsible AI

Cornell University
Machine Learning Foundations · Aug 2025

MY EVERYDAY TOOLBOX
PythonC / C++TypeScript
Scikit-learnPandasNumPyFeature engineeringModel evaluationStatistical analysisData visualization
LLM application developmentAI agent designBrowser automationWorkflow automationREST APIs
Git/GitHubJupyterGoogle ColabClaude CodeVS Code

BEYOND THE SCREEN

Building for people
started before the code.

In 2020, I co-founded Let’s Work for the Kid Haiti. As team lead, I’ve helped organize giving events, coordinate volunteers, and build partnerships to support children in Haiti.

That experience shapes how I work: start with people, listen to what they need, and follow through.

13team members led
$2,000fundraising goal, two fairs
2020the year we started

03 / What’s next

Have something
worth building?

I’m looking for software engineering, AI, and ML internships. If you have a useful problem, an idea, or a team I could contribute to, let’s talk.