Where I’ve worked

AI Research Engineer

Hochschule Offenburg & Hahn-Schickard · Offenburg, Germany

Built evidence-grounded AI systems that turn medical literature into validated probability distributions for clinical Bayesian networks.

• Built a PydanticAI ReAct agent that searches medical literature, reasons over retrieved evidence, and generates validated probability distributions for clinical Bayesian networks.
• Designed a context-tree compression method for large Bayesian CPTs, reducing LLM elicitation calls by 79.4% on average.
• Built a concurrent agent evaluation runtime for benchmarking models and agent variants, with configuration-driven experiments, failure handling, tracing, and automated metrics.

• Healthcare
• Bayesian networks
• Agent evaluation

AI Engineer

GreenEarthX · Palo Alto, CA · Remote

Built agentic compliance and regulatory-reasoning systems for EU renewable-energy regulations.

• Built a LangGraph compliance agent combining multi-step reasoning, RAG, and expert validation, achieving 88% accuracy across 100 compliance scenarios.
• Built a Neo4j knowledge graph from EU regulations, modeling amendments, citations, and cross-document dependencies for multi-hop reasoning.
• Deployed containerized FastAPI services on AWS ECS with Langfuse observability for LLM tracing, token usage, and production debugging.

• Regulatory compliance
• Knowledge graphs
• Agent infrastructure

AI Engineer

Oratio Technologies · Tunis, Tunisia

Built legal-information systems that made regulatory documents faster to ingest, retrieve, and evaluate.

• Reduced average TTFT for a legal RAG system by 67%, from 15s to 5s, by optimizing async execution and introducing pooled, shared clients for MongoDB and Azure Cosmos DB Gremlin API.
• Designed and implemented a CDC and ETL platform provisioned with Terraform on Azure, synchronizing 30,000+ legal documents across MongoDB and Azure Cosmos DB Gremlin API.
• Built synthetic evaluation datasets and an LLM-as-judge framework integrated into CI/CD to detect RAG quality regressions before deployment.

• Legal chatbot
• RAG performance
• Document processing

ML Engineering Intern

ACTIA Engineering Services · Ariana, Tunisia

Worked on low-latency speech-command recognition for automotive systems on resource-constrained microcontrollers.

• Designed, trained, and optimized a CNN speech-command recognition model using TensorFlow for automotive applications.

• Embedded ML
• Automotive systems
• Low-latency inference

Software Developer

IEEE INSAT Robotics and Automation Society Chapter · Part-time · Tunis, Tunisia · On-site

Led development of core software for an autonomous robot that qualified for the Eurobot 2024 international competition in France.

• Designed a decision-making engine with a path planning algorithm fusing LiDAR, computer vision, and odometry for autonomous navigation.
• Engineered a task scheduler optimizing priorities by distance, resource availability, and multi-factor criteria.
• Architected a ROS system with service interfaces uniting LiDAR, camera, navigation, and task scheduling for seamless autonomous operation.
• Established reliable CAN Bus communication via SocketCAN and RS485/CAN HAT between Raspberry Pi 4 and STM32F407.

• Autonomous robotics
• ROS
• Embedded systems