Software Engineer · AI Researcher
I build things that have to survive contact with reality. Then I go looking for where they don't.
CrashLens is a crash-detection platform I co-built that cuts emergency dispatch from roughly 14 minutes to about 30 seconds, validated end-to-end on real hardware, not a demo. SHARP-RAG is a retrieval pipeline I built, then reported honestly when it underperformed its own baseline, because the failure taught me more than a clean win would have. I write about both kinds of work, engineering post-mortems and the personal stuff, when something needs examining instead of shipping. 3.81 GPA, five semesters on the Distinguished List, but I'm more interested in what I do with it: a PhD or an engineering role where the problems have real stakes, not a system that only works in the demo.
IEEE ICCA 2026 paper · 1 preprint · 1 in review cycle
5 projects
English C2 · Arabic Native · French B2
E. (Al)Ghazal, G. Khayat, W. Ishak, B. Farhat, M. Allaw, C. Boustany (advisor) · AUST
Innovation Award, 2nd place, Computer Science track · AUST Engineering & Computer Science Expo 2026
CrashLens is an embedded crash-detection and emergency-dispatch platform. A Raspberry Pi 5 with a six-axis IMU polls acceleration at 100 Hz and confirms a crash by sustained threshold, not vision inference, then dispatches an SOS over LTE on telemetry alone, before any video has uploaded, cutting measured dispatch latency from roughly 14 minutes to about 30 seconds in bench tests. Video and key frames upload afterward on a separate path, triggering a forensic stage that runs Gemini multimodal video analysis in parallel with license-plate OCR, cross-checked against Lebanon's vehicle registry, surfaced through a Flutter driver/responder app and a Next.js insurance dashboard.
Elia (Al)Ghazal · Independent Researcher · Beirut, Lebanon
Multi-hop question answering requires chaining evidence across several documents, a setting in which naive RAG frequently fails because it retrieves once and never verifies whether the retrieved context supports an answer. SHARP-RAG addresses this with a four-agent LangGraph pipeline: a Planner, Retriever, Critic, and Synthesizer cooperate in a cyclic stateful graph where the Critic emits a structured JSON verdict that gates answer generation and drives targeted re-retrieval. Evaluated on 20 HotpotQA fullwiki questions, SHARP-RAG underperforms both baselines.
Finding: the self-correcting loop underperformed naive RAG, driven by over-triggering from the 8B evaluation critic standing in for the intended 70B model, not the architecture itself. Numbers below.
| System | EM | F1 | Latency |
|---|---|---|---|
| Naive RAG | 25.0% | 29.5% | 18.0s |
| Planning Baseline | 25.0% | 28.1% | 24.8s |
| SHARP-RAG v2 | 15.0% | 15.8% | 57.2s |
Title, results, and an interactive demo will be posted here once the review period ends in December 2026.
Honors
Press & talks
Designed and taught a two-day introductory AI/programming curriculum for high-school students, covering programming fundamentals, control flow, and prompting AI systems, with hands-on beginner project building. Built all curriculum and materials solo.
Built backend systems with Spring Boot, including a custom socket server handling ISO-message point-of-sale transactions and Kafka-based messaging.
Open to graduate study discussions, research collaborations, and engineering roles.