Florent Latifi

Software engineer in Prishtina, Kosovo. I build full-stack applications, data pipelines and machine-learning tools, backed by automated tests and CI.

BSc in Computer Science and Engineering from UBT, with MSc studies in Data Science and Artificial Intelligence.

Email me Download CV (PDF)

GitHub LinkedIn

Florent Latifi in a navy UBT graduation gown and cap, arms crossed

Featured project In development

Pragu finds accessibility barriers in Android apps before users hit them

It reads an app the way the TalkBack screen reader does, marks each barrier on a screenshot, and says who it affects and how to fix it. Prag is Albanian for threshold.

  • A Kotlin agent and a Node.js CLI capture any Android app over adb, with no changes to its code.
  • WCAG 2.2 rules check missing names and labels, hidden or stateless controls, text contrast and touch targets.
  • Every rule's precision is measured on labelled screens, and CI fails any change that moves a score.
  • TypeScript
  • Node.js
  • Kotlin
  • Android
  • React Native
  • Jetpack Compose
  • GitHub Actions

Source code on GitHub Technical design

Pragu's report for the cart screen of its React Native test app: the phone screenshot with five numbered markers, next to the first barrier, a 16 by 16 dp touch target, with who it affects, what they experience, and the fix in React Native, Jetpack Compose and Android Views.
A report for a screen recorded from Pragu's React Native test app.

More projects

Gjurmë

Live 2026 · Solo project

Live siteSource on GitHub

Groups Albanian-language news from 19 outlets into stories and shows who covered each one, by perspective, region and ownership.

Gjurmë's dashboard: 1,073 articles today and 9,599 in the last 30 days from 19 outlets, a daily volume chart, and the people and organisations rising in the news.
The live dashboard, in Albanian, the site's default language.
  • Collects RSS feeds, removes duplicates in four layers and groups reports of the same event into stories with TF-IDF clustering tuned for Albanian.
  • Index and query fixes made entity analytics up to 4.5× faster on a year of synthetic data; 500+ automated tests run in CI.
  • Python
  • FastAPI
  • SQLAlchemy
  • PostgreSQL
  • Claude API
  • React
  • TypeScript
  • Docker
  • GitHub Actions
  • Vercel

CineTaste

Live 2026 · Solo project

Live siteSource on GitHub

A movie and TV recommender that learns your taste from your ratings and explains every pick in a sentence you can check.

CineTaste recommending The Green Mile, with its reasons: you rated The Shawshank Redemption highly and it has the same director, Frank Darabont; and it is highly rated but not a chart-topper.
Every pick says why it was chosen.
  • Combines interpretable taste features with vector search in pgvector, so every pick comes with a reason.
  • An offline evaluation on held-out MovieLens ratings guided the ranker: hit rate@20 rose from 0.155 to 0.254.
  • Python
  • FastAPI
  • SQLAlchemy
  • PostgreSQL
  • pgvector
  • Redis
  • React
  • TypeScript
  • Playwright
  • Docker

JavaSmell

BSc thesis 2026 · Bachelor thesis, UBT

Source on GitHub

Finds code smells in Java projects with rules and machine learning, and offers only the refactorings that still compile.

Agreement with expert reviewers, per smell Matthews correlation coefficient, where 1 is perfect agreement.
Show the numbers
Smell Rules and metrics Machine learning Best model
Long Method 0.580 0.724 Random forest
Feature Envy 0.271 0.696 Gradient boosting
Data Class 0.275 0.498 Random forest
Blob 0.232 0.488 Gradient boosting
  • Machine-learning models beat published detection rules on 4,534 expert-labelled samples from 512 Java repositories.
  • Presented at the 15th UBT Annual International Conference on Computer Science and Engineering, October 2026.
  • Python
  • scikit-learn
  • tree-sitter
  • FastAPI
  • React
  • TypeScript
  • Docker
  • GitHub Actions

About me

I graduated in Computer Science and Engineering from UBT in 2026 and went on to an MSc in Data Science and Artificial Intelligence there. My thesis became JavaSmell, which I presented at the 15th UBT Annual International Conference on Computer Science and Engineering. I like work I can measure, so every project here ships with tests and an evaluation.

Education
MSc Computer Science and Engineering (Data Science and Artificial Intelligence), UBT, 2026–present
BSc Computer Science and Engineering, UBT, 2023–2026
Experience
Intern, ProCredit Bank, Kosovo
Based in
Prishtina, Kosovo. Open to remote work.
Languages
Albanian (native), English (B1–B2), Serbian (B1)

Skills

Programming languages
Python, TypeScript, JavaScript, Kotlin, SQL, HTML, CSS
Frontend
React, Vite, TanStack Query, Zustand, React Router, Tailwind CSS, Recharts
Backend
FastAPI, Node.js, SQLAlchemy, Alembic, Pydantic, REST APIs, JWT authentication
Mobile
Android (Kotlin, Jetpack Compose, UiAutomation, adb), React Native (Expo), TalkBack
Data & AI
scikit-learn, NumPy, pgvector, recommender systems, TF-IDF clustering, LLM APIs (Claude), model evaluation
Databases
PostgreSQL, Redis
Tools & DevOps
Git, GitHub Actions, Docker, Gradle, pytest, Vitest, Playwright, axe-core, JUnit, Vercel, Render, Neon
Concepts
layered architecture, CI/CD, automated testing, architecture decision records, query optimisation, accessibility (WCAG 2.2)