Louai Ben Hassine

Louai Ben Hassine

Computer Science student · Université Laval

Second-year CS student who builds systems that keep their intelligence on the machine they run on — retrieval-grounded language models, on-device estimation, and editor tooling. What interests me is the engineering around a model: what it is allowed to see, and what it is allowed to claim.

Open to internships

He follows your cursor. Poke him — but not three times.
T theme · M mute

About

Engineering around the model

I’m in my second year of génie informatique at Université Laval. Most of what I build outside class has the same shape: a system that could call out to a cloud API, and doesn't. A movie recommender that runs a local Llama and is forbidden from recommending anything it didn't retrieve. A nutrition tracker that estimates a plate from a photo in single-digit milliseconds, in a browser tab, with the network panel empty.

The constraint is the interesting part. Cutting off the cloud forces retrieval, grounding and failure modes into the open, where you have to design them rather than hope the model handles it.

I also build tooling inside other people's software — editor plugins for Unreal Engine and scripted panels for After Effects — usually because I hit a manual step in my own workflow that was worth an afternoon to delete.

Selected work

Projects

Three things I built end to end. Each one runs without a paid API behind it.

LBH Cima interface: local llama3.2 selector, prompt box with a Citation toggle, and example queries

LBH Cima · Python · 2026

Local RAG movie recommender

A recommendation engine and chat assistant running entirely on local infrastructure — no API keys, no cloud calls. Queries are embedded with all-MiniLM-L6-v2 and matched against a 4,800-film dataset by cosine similarity; a locally-hosted Llama 3.2 answers through Ollama. The system prompt constrains the model to the retrieved candidates, so it cannot invent a title that isn't in the data. A citations panel exposes the retrieval step and its similarity scores — the part a RAG app usually hides.

  • Python 3.12
  • Ollama
  • sentence-transformers
  • Streamlit
  • scikit-learn
  • NumPy
View repository →
Kalo today view: calories left, macro rings and logged meals Kalo camera view with the framing reticle

Kalo · JavaScript · 2026

On-device calorie & macro tracker

A nutrition tracker that runs entirely in the browser with no dependencies, no build step and no backend — the food table, the photo estimator and the full history live on the device. A photographed plate is reduced to a colour signature and matched against the local food table in roughly 5–10 ms; typed entries like "2 eggs, sourdough toast and a black coffee" parse into three logged items with quantities. Installable as a PWA via a service worker and web manifest.

  • Vanilla JavaScript
  • CSS
  • Service Worker
  • Web Manifest
  • Inline SVG
View repository →

Quoridor · Python · 2025

Board-game engine & API client

An object-oriented implementation of the strategy board game Quoridor, covering game-state management and full rule validation — legal moves, wall placement, and the path-existence constraint that stops a wall from sealing a player in. Ships with a command-line interface, an API client for the course game server, and a unit-test suite. Built as coursework in génie informatique at Université Laval.

  • Python
  • OOP design
  • unittest
  • REST client
View repository →

Notebook

Writing on AI

Notes on where AI is actually going — capability, control, and what the race is really about.

notebook — writing on AI
Latest
Louai Ben Hassine@LouaiBenHassine

AI JUST TOOK ANOTHER BIG STEP FORWARD 🤖

OpenAI has released GPT‑6 Astra, its newest model built for complex professional work. But bigger capabilities bring bigger questions…

#AI #OpenAI #FutureOfWork
Read the full post →

Toolkit

Skills

What I reach for, grouped by what it's actually for.

Languages

  • Python
  • JavaScript
  • C++
  • Swift

AI & retrieval

  • RAG pipelines
  • sentence-transformers
  • Ollama
  • Embeddings
  • scikit-learn

Web

  • Vanilla JS
  • PWAs
  • Service workers
  • CSS
  • Streamlit

Engines & tooling

  • Unreal Engine 5
  • Blender
  • After Effects scripting
  • MCP servers

Practice

  • OOP design
  • Unit testing
  • Git
  • Local-first architecture

Background

Education

Génie informatique — Université Laval

2nd year · Québec City, Canada · Computer & software engineering

Get in touch

Let's build something

Looking for internships, and always happy to talk about retrieval, local models, or tooling. Email is the fastest way to reach me — I read everything.

benhassinelouai620@gmail.com Résumé (PDF)