Hassan Barmandah

Umm Al-Qura University · Makkah, Saudi Arabia

GPA: 3.87 / 4.00

NAMAA Community · Riyadh, Saudi Arabia (Remote)

• Developed end-to-end Arabic NLP systems for LREC 2026 shared tasks, including handwriting recognition (HTR) and diacritization models based on Qwen2.5-VL with LoRA/QLoRA adaptation.
• Designed multimodal pipelines for Arabic manuscript OCR and speech diacritization, integrating Seq2Seq, ASR, and vision-language models under shared-task constraints.
• Built large-scale evaluation and benchmarking framework across 18+ models, performing dialect-level analysis and systematic comparison of Seq2Seq, ASR, and token-classification approaches.
• Engineered data preprocessing and training pipelines for low-resource Arabic dialects, including dataset cleaning, alignment, and format standardization for multimodal inputs.
• Investigated ensemble and decoding strategies including weighted voting, linear + boosting fusion, and normalization techniques to improve robustness on noisy Arabic datasets.

Umm Al-Qura University · Jeddah, Saudi Arabia

• Led development of Haakkim, an arena-style human preference evaluation platform for Arabic LLMs, submitted to IEEE Access.
• Designed large-scale preference learning pipeline using blind pairwise comparisons, Bradley–Terry ranking, and dialect-aware evaluation across Arabic varieties.
• Built scalable data collection and ranking infrastructure supporting 1,100+ comparisons across 67 models, enabling reproducible benchmarking datasets.
• Implemented experimental controls including randomized presentation, balanced sampling, and noise reduction strategies to ensure ranking stability.
• Conducted statistical validation and analysis of ranking reliability across models, supporting research-grade evaluation and reproducibility standards.

Synapse · Jeddah, Saudi Arabia · Deputy Director Mar 2026–Present · AI Engineer Oct 2025–Mar 2026

• Coordinate development of web applications across technical teams, ensuring alignment with project requirements and delivery goals.
• Developed Retrieval-Augmented Generation (RAG) systems integrating vector databases, embedding models, and LLM inference pipelines for internal applications.
• Built document ingestion and retrieval workflows for knowledge-based AI systems, including chunking, embedding generation, and semantic search optimization.
• Integrated AI modules with backend services to support functional features in internal applications and improve system usability and response quality.

Lucidya · Riyadh, Saudi Arabia (Remote)

• Built production-grade agentic system for Confluence and codebase using RAG and AST-based graph representations, deployed in production for 30+ engineers.
• Designed multi-stage LLM agent pipeline enforcing structured outputs, hallucination mitigation strategies, and automated quality scoring mechanisms for reliability.
• Engineered backend AI infrastructure using Celery async orchestration, FAISS vector indexing, hybrid retrieval pipelines, and LLM streaming via AWS Bedrock.
• Built observability and monitoring layer tracking usage, latency, cost, and feedback signals across all LLM-based agent workflows in production.
• Developed internal tooling and integration layer for enterprise workflows, enabling seamless interaction between agents, APIs, and engineering systems.

Uptrail · London, UK (Remote)

• Performed exploratory and diagnostic data analysis using Python, SQL, and Tableau to clean, transform, and validate datasets for reporting and business insights.
• Designed and maintained dashboards and interactive visual reports to track KPIs and support client-facing decision-making processes.
• Translated raw business requirements into structured analytical queries and reporting pipelines for stakeholder use.

Recommendation Letter

Smart Methods · Jeddah, Saudi Arabia (On-site)

• Developed robotic prototypes integrating embedded systems (Arduino/ESP32), sensors, and actuators for real-time automation tasks.
• Built computer vision pipelines using OpenCV for object detection, classification, and perception-based robotics applications.
• Implemented ROS2-based robotics workflows and integrated speech (STT/TTS) and LLM components for interactive system behavior.
• Designed and developed web-based interfaces for robotic control, monitoring, and user interaction across robotics, IoT, AI, and software development tasks.