- Designed and built a fully offline retrieval-augmented generation (RAG) system (React 19 + Vite frontend, Python FastAPI backend, Ollama runtime) over a curated 136-station UAE EV charging dataset from the Open Charge Map API — zero cloud API calls, all inference on-device on an NVIDIA RTX 4070. - Implemented a word-boundary regex semantic query parser (6-dimension filters: speed, operator, connector type, power, city, exclusion) with Haversine geolocation for nearest-charger retrieval. - Designed a two-phase anti-hallucination pipeline (intent parsing → backend execution → grounded narration), eliminating hard hallucinations across all 80 evaluation runs. - Benchmarked Llama 3, Mistral 7B, Gemma 7B, and Phi-3 — Llama 3 best overall: 93.3% retrieval accuracy, 87.6% reasoning, 97.5% tool use at 5.9 s latency.

- Led the design and development of a grid-based game pitting a human player against an AI agent across resource-collection objectives, with paired gem mechanics and per-agent collection logic. - Implemented and tuned A* (with admissible heuristic and corrected cost function) and BFS pathfinding for the AI agent over a structured state space. - Authored the technical LaTeX report covering algorithm analysis, evaluation methodology, and an ethics chapter on AI competitive systems.

Spring 2025

Spring 2023

100% tuition (highly competitive UAE merit award).

(high school graduating class).