Abdulrhman Mohammed

AI knowledge retrieval system. A secure RAG platform with hybrid BM25/vector search, reciprocal rank fusion, cross-encoder reranking, and citation-grounded LLM responses. Role-based document access, FastAPI services, and automated evaluation pipelines measuring retrieval accuracy, answer faithfulness, latency, and cost.

Autonomous AI workflow system. An agent that runs multi-step customer support workflows through function calling, external API integrations, and persistent state management, with human-in-the-loop approval, tool execution safeguards, and automated evaluations of task completion, reliability, and operational safety.

AI document processing platform. Combines OCR, multimodal LLMs, and structured extraction to turn Arabic and English invoices, receipts, and PDFs into validated JSON. Asynchronous workflows with background workers, Pydantic validation, PostgreSQL storage, and human review for uncertain extractions.

Fine-tuned open-weight language models with LoRA/QLoRA, Hugging Face Transformers, and PEFT, benchmarked against baseline models. Deployed optimized inference services with vLLM and FastAPI, evaluating model quality, GPU memory use, throughput, and latency.

An AI-powered study assistant for interactive learning and question answering through LLM integration, built end to end.

Bilingual Arabic/English platform for the Qassim date industry: product catalog, producer and trader directories, exhibitions, an AI assistant, and analytics.