// ♞ 2× National Chess Champion · INSAT — Computer & Networks Engineering (RT) · Class of 2027

HAZEMBENALAYA

I build agentic AI systems

Fourth-year engineering student at INSAT (Tunis) and two-time national chess champion, turning large language models into dependable, on-premise software — retrieval, agents, forecasting — with the back ends to run them. The board taught me to calculate a few moves ahead; I build software the same way.

See the experience ↓ GitHub ↗

TUNIS, TUNISIA hazem.benalaya@insat.ucar.tn LINKEDIN/HAZEM-B-5BA743288

01 ♟

About

I'm Hazem Ben Alaya, a Computer & Networks engineering student at INSAT. I like the unglamorous middle of AI products: clean datasets, grounded retrieval, measurable models — and the APIs that behave when everything else is on fire.

Over the last two years I've shipped a RAG knowledge assistant for internal support teams, agentic test generators that turn API specs into executable pytest suites, and an on-premise analytics co-pilot for Poulina Group Holding's sales controllers. When I'm not orchestrating LLMs, I'm breaking (and patching) things with Securinets INSAT in CTF competitions or building team projects with Junior Entreprise INSAT.

Before code, there was chess: national champion, 1st place two years running. The board taught me calculation, pattern recognition and nerves — evaluate the position, pick the strongest move, commit. I debug a misbehaving model exactly the same way.

0National chess titles · back-to-back

0Internships completed

0Place · Darkest Hour CTF '24

0INSAT class of

02 ♞

Experience

Aug — Sep 2026

Agentic AI Intern — Sales Control Co-pilot

Poulina Group Holding · Tunis

Built an on-premise conversational co-pilot for sales management control: a LangGraph intent router orchestrating Pareto / ABC×XYZ analytics and forecasting tools over multi-year, multi-entity sales history.
Shipped a LightGBM forecasting module with P10 / P50 / P90 landing scenarios, tuned to Tunisian market seasonality (Hegirian calendar, Ramadan, Eid).
Added a RAG layer over management-control procedures and a React chat UI with interactive Plotly visuals and PDF export — FastAPI, PostgreSQL, Docker, local Ollama inference.

Jun — Jul 2026

AI Engineering Intern — Agentic Test Generation

Avidea · Agentic quality engineering

Developed a LangGraph agent that turns OpenAPI / Swagger specs and Postman collections into executable pytest suites — happy-path, negative and boundary scenarios with concrete request/response assertions.
Closed the loop with a coverage audit, live execution against running APIs, and one constrained LLM rewrite pass to repair failing tests.
Built the companion SwaggerToTest engine: language-neutral test plans rendered deterministically to pytest and Karate, with a fully offline mode.

Jul — Sep 2025

AI Intern

Next Step IT · Applied AI

Designed a local, secure RAG chatbot giving internal technical staff contextual, source-grounded answers — built to boost knowledge access and onboarding efficiency.
Built a FAISS semantic search engine integrated with a local LLM via LangChain, behind a React interface.
Contributed to an automated document-generation module (audits, configs, HLD/LLD).

2023 — Now

Active Member — Cybersecurity

Securinets INSAT · CTF Team

Competed in national and international Capture The Flag challenges; 4th place out of 50+ teams at Darkest Hour CTF 2024.

2 years running

National Chess Champion

National championship · 1st place, two consecutive years

Took the national title back-to-back — preparation, calculation and keeping a clear head when the position turns worse.
The habits transfer directly to engineering: pattern recognition, thinking several moves ahead, and playing the position — not the piece.

03 ♜

Selected Work

P·01↗

Agentic Test Generator

An LLM-orchestrated LangGraph workflow that converts OpenAPI specs and Postman collections into executable, coverage-audited pytest suites — then reruns and rewrites failures.

Python
LangGraph
pytest
OpenAPI

P·02↗

SwaggerToTest

OpenAPI YAML/JSON in, runnable contract tests out — deterministic offline generation with a language-neutral plan layer targeting pytest and Karate.

OpenAPI
pytest
Karate
FastAPI

P·03↗

Spam Detection System

Fine-tuned BERT + scikit-learn pipeline reaching 95% classification accuracy, served over FastAPI with a React inbox UI and Spring Boot integration.

BERT
scikit-learn
FastAPI
React

P·04↗

Power Grid Optimization

Mixed-integer linear programming with GurobiPy minimizing generation and transmission costs under city-demand constraints — Flask REST API + Electron desktop app.

GurobiPy
MILP
Flask
Electron

P·05↗

MMCows · Behaviour Recognition

Extended the NeurIPS 2024 MMCows benchmark on a 4-person research team — multi-rate sensor synchronisation, sliding windows and domain-informed time/frequency features with leave-2-cow-out validation. Macro F1 0.82, past the published sensor baseline.

PyTorch
LSTM
Transformer
XGBoost

More experiments on GitHub

ML notebooks, Spring Boot & Java builds, a JS robot, and everything in between.

github.com/hazemBA20 ↗

04 ♛

The Pieces

♞Knight · Languages

Python
Java
TypeScript
JavaScript
C
C++
PHP

♛Queen · AI & Machine Learning

LangGraph
LangChain
RAG
FAISS
Ollama
PyTorch
scikit-learn
LightGBM
Transformers

♜Rook · Web & Back end

FastAPI
React
Node.js
Spring Boot
Symfony
Laravel
Plotly

♝Bishop · Data & DevOps

PostgreSQL
MySQL
MongoDB
Qdrant
Docker
Git
GitHub

// Open to PFE internships & junior AI/software roles

Your move→

Have a dataset nobody can query, an API nobody can test, or a model nobody trusts? That's my kind of problem — tell me about it and we'll find the strongest continuation.

Email me → LinkedIn ↗ GitHub ↗