Hi. I am

Taha Bouhafa

> AI Engineering Student

Exploring intelligent systems at the intersection of vision and data.
I specialize in Computer Vision and Deep Learning, building robust models
that turn visual data into accurate and impactful solutions.

Download Resume

About Me

I’m Taha Bouhafa, an engineering student specializing in Big Data and Artificial Intelligence at ENSA Tétouan. Passionate about AI, I focus on deep learning, natural language processing, and computer vision.

With practical experience in Python, PyTorch, TensorFlow, and scikit-learn, I enjoy applying state-of-the-art techniques including Retrieval-Augmented Generation (RAG) and LangChain to build intelligent, context-aware applications.

I am eager to grow as an AI practitioner, contribute to impactful projects, and solve real-world problems by combining data, language, and intelligence.

Tech Stack:

Programming & AI:

Python,PyTorch,TensorFlow,Keras,scikit-learn,NumPy,Pandas,OpenCV,LangChain,RAG

Tools & Platforms:

Jupyter Notebook,Flask,Streamlit,Docker,Git,SQL

Education

ENSA Tétouan, Morocco

Sept 2023 – Present

Engineering Cycle in Big Data and Artificial Intelligence. Courses: Machine Learning, NLP, Deep Learning, Computer Vision, Big Data, Data Visualisation .

ENSA Tétouan, Morocco

Sept 2021 – July 2023

Integrated Preparatory Classes. Strong foundations in mathematics, physics, and computer science.

EST Agadir, Morocco

Sept 2020 – July 2021

DUT in Computer Science. Focus on programming, algorithms, databases, and software development.

Morocco

Sept 2019 – July 2020

Baccalaureate in Mathematical Sciences with honors.

Experience

2025

CRI Tangier-Tétouan-Al Hoceima

June 2025 - September 2025

AI Research Internship

During the internship I explored Retrieval‑Augmented Generation (RAG) architectures and evaluated a range of open‑source large language models (Mistral, DeepSeek, LLaMA). I built a complete RAG pipeline with LangChain, integrating vector‑based retrieval (FAISS), keyword‑based retrieval (BM25), and contextual generation through a FastAPI backend. I also created a responsive web front‑end (HTML/CSS/JS) that interfaces with the API, delivering an interactive user experience designed to support the information‑retrieval needs of Regional Center of Investments, Tanger‑Tetouan‑Al Hoceima Region (CRI TTAH). The system was evaluated with RAGAS to assess its retrieval and generation performance.

Adventures in Development:
Top Personal Projects

U-Net3+ with Deep Supervision — Aerial Image Segmentation (Potsdam)

Semantic segmentation of ultra-high-resolution aerial RGB images (ISPRS Potsdam) using U-Net3+ with optional deep supervision. Includes tiling pipeline, RGB→class mask conversion, training scripts, metrics (IoU, Dice), and visualization.

Features

U-Net3+ architecture with full-scale skip connections

Optional deep supervision for intermediate decoder outputs

Tiling pipeline for 6000×6000 Potsdam images (256/512 px)

RGB mask → indexed class mask converter

Evaluation with IoU and Dice Score, plus visualizations

Tech Stack: PyTorch, TorchMetrics, NumPy, Pillow, Matplotlib

View project

U-Net3+ with Deep Supervision — Aerial Image Segmentation (Potsdam)

CivicVision, Pedestrian Crossing Behavior Recognition

AI-based computer vision system for predicting pedestrian crossing intention in real time. Uses a Two-Stream I3D architecture to analyze RGB and optical flow data, enabling risk-aware decision making for intelligent transportation systems and autonomous vehicles.

Features

Two-Stream I3D model using RGB and optical flow inputs

Spatio-temporal behavior analysis including motion and body cues

Real-time crossing intention prediction with risk probability score

Decision engine for trajectory maintenance or emergency braking

Interactive React dashboard with live metrics and visualization

Tech Stack: Python, TensorFlow, Keras, I3D, OpenCV, React, JAAD Dataset

View project

CivicVision, Pedestrian Crossing Behavior Recognition

Visual Question Answering (VQA)

Developed a deep learning system that answers natural language questions about images. Combined ResNet50 for image encoding and BERT for text processing to predict answers from the VQA v2.0 dataset.

Features

Integrated BERT and ResNet for multi-modal input processing

Processed COCO-VQA JSON annotations for majority answers

Tokenized questions and created a top-1000 answer vocabulary

Extracted visual features from images using ResNet50

Tech Stack: PyTorch, Transformers, ResNet50, BERT, NumPy, Matplotlib

View project

ODQA, Open Domain Question Answering System

End to end open domain question answering system that retrieves relevant passages from a large document collection and extracts precise answer spans using a LoRA fine tuned BERT reader. Includes a full stack web application for interactive querying.

Features

Dense Passage Retrieval using DPR and FAISS indexing

LoRA adapted BERT reader for efficient answer span extraction

Confidence scoring and top k passage selection

FastAPI backend with question and conversation endpoints

Interactive React web interface with authentication support

Tech Stack: PyTorch, Hugging Face Transformers, FAISS, FastAPI, React, Tailwind CSS

View project

ODQA, Open Domain Question Answering System

Sign Language Recognition (ASL)

Built a real-time web app for American Sign Language detection using YOLOv8. Users can upload images, videos, or use live webcam stream to detect ASL letters instantly.

Features

YOLOv8-based object detection

Real-time webcam detection with OpenCV

Video conversion using MoviePy

Flask app with multi-page routing (Home, Detect, Stream)

Tech Stack: Python, Flask, YOLOv8, OpenCV, MoviePy, Bootstrap

View project

Semantic Book Recommender

Semantic recommendation system using sentence-transformers and LangChain. Users get book suggestions based on prompts, emotions in descriptions, and filtered categories.

Features

Semantic search with MiniLM embeddings

Emotion-based sorting (joy, fear, sadness, etc.)

Category filtering

Interactive Gradio dashboard

Tech Stack: LangChain, ChromaDb, Gradio, MiniLM, Pandas, NumPy

View project

Skill Radar: AI & Data Job Forecasting

Dashboard and prediction platform providing insights on Data Science & AI job trends, skill demand forecasting, skill recommendations, and salary estimations based on real job data.