Who I Am
Hi, I'm Khadija Kriaa
I'm a final-year Data Engineering & Decision Systems (IDSD) student at ENET’Com, Sfax, specialising in Data Science. I’m now looking for a 6-month end-of-studies internship (PFE) starting February 2027.
My work sits at the intersection of AI engineering and software craft — I build LLM applications with RAG pipelines and agentic orchestration in Python, then back them with solid REST APIs and relational data models. I'm driven by curiosity, drawn to agentic systems and software reliability, and motivated by shipping AI into real business products.
Beyond code, you'll find me playing guitar, diving into tech or self-improvement books, or watching a thought-provoking sci-fi series. I believe the best engineers are also the most well-rounded humans.
Core Focus Areas
Featured Projects
AI Assistant for Banking Circulars (RAG)
An internal assistant that answers staff questions about the bank's circulars in French, grounded in the retrieved documents and citing the exact circular and article.
Mainstream PDF extractors could not decode the circulars' embedded fonts, so I wrote a custom extractor, then built article-aware chunking, BGE-M3 embeddings in Qdrant, hybrid dense + BM25 retrieval fused with RRF, and a LangGraph retrieve-then-generate flow behind a Streamlit interface. Every design choice was backed by an evaluation harness rather than intuition.
Key Points
- Compared 13 chunking strategies and 2 embedding models (MRR, Recall@K)
- Benchmarked candidate LLMs on evidence, citations and latency
- Refuses questions about circulars that do not exist
- Answers cite the source circular and article
Tunisian Real-Estate Price Prediction
An end-to-end pipeline that collects Tunisian property listings, cleans them and predicts sale prices in Tunisian dinars to support property revaluation.
Wrote scrapers for Tayara.tn and NewKey, cleaned the raw listings into a sale-only dataset, engineered features from structured fields and from listing descriptions (TF-IDF, amenity flags), then compared seven regression models in reproducible scikit-learn pipelines tuned with RandomizedSearchCV.
Key Points
- Scrapers for Tayara.tn and NewKey
- Text features from listing descriptions
- Leakage-free, reproducible pipelines
- XGBoost, Gradient Boosting and LightGBM all above R² 0.71
Tunisian Dialect Sentiment Classification
A binary sentiment classifier for Tunisian dialect tweets written in both Arabic script and Arabizi (Latin script), fine-tuned from MARBERT.
Built the preprocessing pipeline, including a CODA-style Arabizi-to-Arabic transliteration step, merged both scripts to double the training data, then fine-tuned MARBERT on Hugging Face and compared it against logistic regression and a linear SVM trained on frozen MARBERT embeddings.
Key Points
- Handles both Arabic script and Arabizi
- Fine-tuning beat frozen-embedding baselines (78% accuracy)
- Stratified train / validation / test split
- Evaluated on 9,559 held-out tweets
Hybrid Agent for Driving Assistance
A hybrid multi-agent system that detects road situations in real time using VEINS simulator data and generates natural language + textual alerts to assist drivers safely.
Combines a reactive safety agent with a deliberative BDI agent for contextual reasoning, forming a true hybrid multi-agent architecture. Multiple independent ML agents (overtaking, intersection, dangerous zones) collaborate by feeding beliefs into the BDI layer, which generates context-aware alerts via natural language generation and text-to-speech.
Key Points
- Real-time road situation detection
- Natural language alert generation
- Hybrid multi-agent architecture
Skills & Expertise
Soft Skills
Also Familiar With
Internships
AI & Data Intern — RAG, LLM & Agentic AI
Union Internationale de Banques (UIB)
Built an internal assistant that lets staff query banking circulars in natural language, grounded in retrieved source documents and citing the circular and article behind every answer.
- Developed a Python RAG pipeline covering PDF extraction, chunking, embeddings, retrieval and LLM generation
- Orchestrated the pipeline with LangGraph for agentic control flow
- Implemented hybrid retrieval combining BGE-M3, BM25 and Qdrant, fused with Reciprocal Rank Fusion (RRF)
- Built an evaluation harness (282 questions, MRR and Recall@K) to compare 13 chunking strategies and 2 embedding models
- Benchmarked candidate LLMs: 0.87 citation precision and 15/15 correct refusals on questions about non-existent circulars
- Built a Streamlit interface surfacing the sources used for every generated answer
Data Science Intern — Data Pipelines & Predictive Modeling
Société Tunisienne de Banque (STB)
Built a price prediction pipeline for Tunisian property listings, from web scraping to systematic model comparison, to support real-estate revaluation.
- Scraped 16,000+ listings from Tayara.tn and NewKey and cleaned them into 4,413 sale listings (Python, Pandas, NumPy)
- Engineered features from structured fields and listing descriptions (TF-IDF, amenity flags)
- Built reproducible pipelines with Scikit-learn Pipeline, ColumnTransformer and RobustScaler
- Compared 7 regression models; best: XGBoost with R² = 0.72 on the held-out test set
- Applied cross-validation and RandomizedSearchCV for hyperparameter tuning
Software & AI Intern — Back-End, REST APIs & Document Processing
Coconsult Nearshore
Built a document processing platform combining Tesseract OCR with the Cohere API to extract and structure information from PDFs and images.
- Combined Tesseract OCR and the Cohere API to extract and structure data from PDF and image documents
- Developed REST APIs with Spring Boot and designed the relational storage layer under PostgreSQL
- Integrated JWT-based authentication with Spring Security for secure data exchange
- Contributed to the development of the Angular frontend
Education
Engineering Degree — Data Engineering & Decision Systems (IDSD)
ENET'Com — Ecole Nationale d'Electronique et des Telecommunications de SfaxFinal-year engineering programme specialising in Data Science. Curriculum covers advanced algorithms, databases, machine learning, AI engineering, and system architecture.
Mathematics & Physics (MP) preparatory track for the National Engineering Entrance Competition (Concours National d'Ingenieurs). Passed the national competition to qualify for elite engineering programmes.
Get In Touch
Open to Opportunities
Internships, research, collaborations
I’m a final-year Data Engineering & Decision Systems student at ENET’Com, specialising in Data Science. I’m looking for a 6-month end-of-studies internship (PFE) starting February 2027. If you have an opportunity or just want to chat, feel free to reach out!