PFE intern, data or AI
I built and run HiresHere: a daily pipeline over about 540 companies' job feeds, with an AI classifier and a verdict for 243 countries.hireshere.app, live since October 2026
Software & data engineering student. I built and run HiresHere, a live job site with a data pipeline and an AI classifier behind it.
Open for a PFE in data or AI, 2027
I built and run HiresHere: a daily pipeline over about 540 companies' job feeds, with an AI classifier and a verdict for 243 countries.hireshere.app, live since October 2026
A remote job site that tells you which jobs you can actually apply for from your country, and quotes the line in the posting that proves it.
11,566 jobs listed · 597 companies · 243 countries · $1.85 to classify 4,563 postings
Jobs, companies and levels come from the site's own search index after that day's run; the cost is Gemini's batch estimate for the 4,563 postings classified the same day.
A citizen photographs a pothole, a YOLOv8n model grades its severity, two reports within 5 m merge into one, and a RAG agent drafts the repair order citing the actual law, routed to the municipality or the roads authority.

A multi-agent assistant for my faculty (FSB) that answers from the faculty's own documents and shows its sources. Hybrid retrieval over pgvector and Postgres full-text, fused with reciprocal rank fusion.

Student portal for FSB: GPA and honours on the Tunisian 20-point scale, timetables, PDF documents, and a Gemini assistant served through Supabase Edge Functions.

An Airflow DAG over seven public datasets (crash reports, EV registrations, air traffic, GTFS transit, EPA facilities, car sales): extracted, staged, cleaned with pandas, quality-checked and loaded into Postgres.
A Flutter app: photograph a leaf or an insect and get one of 40 North African crop pests identified by Gemini, with treatment advice in French, English or Arabic, and a scan history kept offline.

One of the 40 pests in the app's catalogue (citrus leafminer). Photo: Franck A., CIRAD.
FSB Nexus, the academic assistant above, built for my faculty.
An annotated set and a 100-question set, scored on hit@5, MRR, recall@10, answer rate and wrong-answer rate. The jump came from finding a vector-index misconfiguration, then adding hybrid search (vector plus full-text, reciprocal rank fusion).
Automated multichannel customer journeys (email, SMS, push) in Dynamics 365 Customer Insights, and wired their data into central reporting with Power Automate and Microsoft Fabric.
I'm a new open source contributor. I started with the tools and data I rely on for HiresHere: reporting and fixing problems in company directories, job-board datasets and AI frameworks, and sharing what running them in production taught me.
SQL, PostgreSQL + pgvector, Airflow, pandas, SQLite, MongoDB, Supabase, data quality checks
RAG, hybrid search, evaluation harnesses, Gemini, Mistral, sentence-transformers, scikit-learn, YOLOv8
Python, TypeScript, FastAPI, Pydantic, Astro, React, Flutter, Docker, GitHub Actions, Cloudflare