Hybrid RAG + Text-to-SQL System for Mixed-Type Data
Designed a hybrid architecture for AI chatbots over knowledge sources containing both numeric and free-text columns — routing each column to the retrieval method best suited to its content. Numeric columns are handled via SQL querying; free-text columns via semantic search — combined within a single query pipeline. Implemented on PostgreSQL.
RAG
Text-to-SQL
PostgreSQL
LLM
Semantic Search
Arabic ASR Data Research — Egyptian Dialect
Surveyed major datasets for Egyptian Arabic ASR (Common Voice, MGB-2/3, FLEURS, CALLHOME, ArZen, and others), comparing WER results across existing models to identify high-performing data sources. Contributed to validating newly collected data for the project.
Arabic NLP
ASR
Whisper
Research
PDF Intelligent Assistant
RAG-based chatbot for PDF question-answering using advanced LLMs. Allows users to upload documents and query their content conversationally.
RAG
LLM
Chatbot
AI Customer Service Chatbot
OpenAI-powered automated response system built with prompt engineering for consistent, context-aware customer interactions.
OpenAI API
Prompt Engineering
Chatbot
Audio Classification: AST vs. HuBERT
Fine-tuned and benchmarked Audio Spectrogram Transformer and HuBERT on the Speech Commands dataset to compare model performance on audio classification.
HuggingFace
Audio ML
Transformers
AI Personal Trainer
Real-time workout form analysis and repetition counting using computer vision and pose estimation.
Computer Vision
OpenCV
Real-time
Gesture Volume Control
Hand landmark detection for system-level volume control, using computer vision to translate gestures into OS-level audio commands.
Computer Vision
MediaPipe
OpenCV