The team

Six people, six modules

Verify was built by a team of six 3rd-year AI engineering students at ESPRIT School of Engineering. Each module is led by a dedicated specialist; together we combine computer vision, NLP, forensics and regulatory knowledge into one platform.

The Verify team at ESPRIT
The Verify team — ESPRIT 2025-2026
Class 3IA1 · Group FBAIS · Tutors: Mme Sonia Mesbah & M. Mustapha Trabelsi
Yassmine Nouisser · Islem Tellili · Malek Tirellil · Youssef Jouini · Rayen Ayat · Maryem Ouichka
Y

Yassmine Nouisser

Project Lead — Cosmetic Ads Fact-Check & Platform Integration

Owns the cosmetic-ads fact-checking module (Whisper + EasyOCR + a curated EU-655/2013 knowledge base with 25 patterns, 33 fake claims and 145 verified ingredients). Coordinates the full Verify platform — frontend, FastAPI backend, the Verify Assistant chatbot, and the HuggingFace Space deployment.

I

Islem Tellili

Lead — Fake Media Detection (deepfake & AI-generated)

Trains and validates the ResNet50 ensemble that powers face-deepfake detection, paired with an MTCNN face detector and a two-stage AI-image consensus to catch GAN portraits and diffusion-model generations the deepfake model alone would miss.

M

Malek Tirellil

Lead — Visual Manipulation & Persuasion

Designs the multi-signal pipeline that scores persuasion in an image: TrOCR for on-screen text, a DistilRoBERTa classifier fine-tuned for clickbait, CLIP zero-shot for sensational visual style, and a curated OCR-driven urgency lexicon in French and English.

Y

Youssef Jouini

Lead — Image-Caption Coherence

Builds the cross-modal coherence pipeline that compares an image (or a video) with the caption it is published with — fusing CLIP ViT-L/14, YOLOv8m object detection, EasyOCR on-screen text, SAM segmentation and Whisper audio transcription into a single verdict.

R

Rayen Ayat

Lead — Image Tampering & Photoshop Forensics

Owns the photo-forensics module. The pipeline runs three classic model-free techniques — Error-Level Analysis, noise residual and JPEG ghost — to localise edits (splicing, inpainting, copy-move) on the image with no trained classifier in the loop.

M

Maryem Ouichka

Cross-Modal Coherence Analysis

Investigates how the five detection signals agree (or disagree) on the same media — a transverse study that feeds calibration data back into the coherence and caption-fidelity modules and helps the newsroom interpret borderline verdicts.

Our mission

Provide the public, journalists and Tunisian institutions with an independent visual verification toolkit, transparent on its methods and free from any commercial interest in steering its verdicts.

Want to join us?

We're hiring an engineer and a fact-checking journalist.

View open positions