HASSAN AQRABAWI / DUBAI
APPLIED AI ENGINEER
I build reliable AI systems and useful products—from multi-LLM consensus and RAG to computer vision, intelligent automation, and production applications.
VIEW WORK[002] / SELECTED WORK
Systems built for the real world.
Six projects — shipped products and active builds, from edge vision to operational AI.
[003]
Client systems, AI training, and production engineering.
4Square Capital
I build APIs, model integrations, and answer-verification tools for Genius², the AI engine behind AskDiana. I also work with open-source models and fine-tuning.
iEnglish Languages Institute
I taught machine learning fundamentals, prompt engineering, and practical automation with AI tools.
Innodata
My freelance work covers data annotation and prompt engineering: labeling data and writing prompts for AI tasks.
PHDC
As the center's sole AI engineer, I built CRM workflows, a document-based AI assistant, and internal tools for administrative work. I continue part-time. Clinical decisions stay with qualified staff.
Topaz
I built tools to identify speakers in customer-service calls and analyze each speaker's sentiment.
[004]
Based in Dubai, I work with local model hosting, training, and fine-tuning, and build AI-powered apps and APIs. I'm familiar with LLMs, RAG, AI agents, and computer vision. At 4Square Capital, I'm responsible for the Genius² model.
EVERYTHING ELSE I'VE BUILT
[005] / LAB
Conversation intelligence for customer service: speaker diarization, per-speaker sentiment, and call analytics.
pyannote / Whisper / Sentiment
Conversation intelligence for customer service: speaker diarization, per-speaker sentiment, and call analytics.
Document chat with no vector database — PageIndex reads the full PDF, no chunking, no embeddings. Fewer moving parts, faster iteration.
PageIndex / Streamlit
Document chat with no vector database — PageIndex reads the full PDF, no chunking, no embeddings. Fewer moving parts, faster iteration.
Chest X-ray pneumonia classifier run as a proper MLOps loop — tracked with MLflow, tuned with Optuna, and served.
PyTorch / MLflow / Optuna
Chest X-ray pneumonia classifier run as a proper MLOps loop — tracked with MLflow, tuned with Optuna, and served.
Evidence-first debugging skill for OpenAI Codex: reproduce, isolate with a focused test, apply the smallest fix, verify. Open source, run with npx codex-debugger.
Node.js / Codex / npm
Evidence-first debugging skill for OpenAI Codex: reproduce, isolate with a focused test, apply the smallest fix, verify. Open source, run with npx codex-debugger.
Full website build for Dubai's largest premium fitness & wellness facility — cinematic scroll, dark editorial layout, freelance client work.
Web / Freelance / Frontend
Full website build for Dubai's largest premium fitness & wellness facility — cinematic scroll, dark editorial layout, freelance client work.
Inspired by Leen Kasir. I had to bring probability to a Battleship fight just to beat her. A Python + Tkinter assistant that maps the odds and picks the next shot. Totally normal levels of competitiveness.
Python / Tkinter / Probability
Inspired by Leen Kasir. I had to bring probability to a Battleship fight just to beat her. A Python + Tkinter assistant that maps the odds and picks the next shot. Totally normal levels of competitiveness.
Financial fact-verification built on Quorum's consensus core — take a claim through multi-LLM consensus, then verify it against a data pipeline. Parked deliberately.
Quorum / Finance
Financial fact-verification built on Quorum's consensus core — take a claim through multi-LLM consensus, then verify it against a data pipeline. Parked deliberately.
Post-discharge patient monitoring over WhatsApp, designed around Malaffi — the UAE's national health record. Red Bull Basement UAE entry.
WhatsApp API / LLM triage
Post-discharge patient monitoring over WhatsApp, designed around Malaffi — the UAE's national health record. Red Bull Basement UAE entry.
MAIN SIGNAL RESUMES