Student Drive MultiModal RAG Assistant
Architected an event-driven Multimodal RAG system utilizing a FastAPI Ingestion Engine and an independent Chat Inference Node to transform unstructured academic data (long video lectures, handwritten scribbles, complex PDFs and images) into a highly accurate, citation-backed knowledge base.
- Engineered a CUDA-optimized ETL worker swarm that extracts text via a hybrid RapidOCR/PaddleOCR stack, refines accuracy using Gemini-2.5 for post-OCR correction, and leverages a Florence-2 VLM and YOLOv8 hybrid for semantic layout analysis and diagram captioning.
- Processed complex media by transcribing audio via Whisper v3 with speaker diarization and reconstructing visual data grids into LLM-readable Markdowns using Microsoft Table Transformers.
- Designed the cognitive reasoning core using a LangGraph cyclic state machine for dynamic query routing, generating high-dimensional embeddings and mapping them into ChromaDB to cite answers.
- Automated the entire ingestion lifecycle utilizing n8n webhooks to instantly synchronize Google Drive uploads with the backend FastAPI inference nodes.
- Built real-time administrative tools, empowering Teaching Assistants to expand the curriculum with YouTube transcripts and surgically recalculate and update specific vector chunks dynamically through the chat interface.
Tactic Zone – Match Analysis & Coach Assistant
Innovated software aimed at improving match preparation. The tools streamline match analysis and coaching by replacing manual work with automation.
- AI-Powered Match Analysis: Annotated a new dataset using CVAT and RoboFlow, used YOLOv8X with some integrated modules for detailed feature extraction and strategic insights (28 feature/player) and (36 feature/team).
- Smart Recommendations Engine: engineered a recommendation system that analyzes 900+ matches to suggest optimal formations, lineups, and substitutions.
- Opponent Analysis: Integrated LLM for detailed match reports and counter-strategies.
- Training Session Planning: Used all these outputs to generate tailored sessions based on insights and player needs.
- Interactive Chatbot: Integrated this Data and prompt-tuned Gemini LLM for real-time queries and adjustments.
Speaking Glove – Real-Time Sign Language Translator
Developed a solution that integrates embedded systems and AI to translate sign language into text and speech.
- Integrated a KNN algorithm for mapping flex and tilt sensor readings to words.
- Handled input/output such as tilt and flex sensors, sound modules, LCD, and more.
- logic to interpret and reset sentences based on stopping words.
- Efficient Scheduling: Managed parallel task scheduling using timers and flags for smooth processing.
Smart Home Temperature Control System
Designed a temperature control system with both manual and automatic modes based on RTOS. The system utilized two AtMega32 microcontrollers connected via SPI protocol.
- Control Unit: Handles temperature reading, conversion, threshold selection via keypad, and LCD.
- Manual Unit: Controls heater, fan, and manual switches through external interrupts.
Clinic Management System – C and MIPS Assembly Versions
Built a complete software having admin and doctor modes to facilitate reservations and manage patients' records.
The project is deployed in C, but I also constructed a MIPS version of it from scratch as training on assembly for future debugging projects.
Already Washed – Flutter App
Applied a mobile app to streamline laundry services, enabling users to manage orders efficiently. Key features included:
- User authentication: login, registration, and secure password recovery.
- Order Management: real-time tracking, status updates, and order placement.
- Membership & Subscription: Manage packages and subscriptions.
- UI/UX: Designed user-friendly navigation and intuitive interfaces.
- State Management: Used Cubit for predictable state management.
- Firebase Integration: Handled data storage, email verification, and notifications for real-time updates.
HDRNI – Flutter App
A mobile app for automating student attendance, courses' system and grading at Shoubra Faculty of Engineering. Project highlights:
- Built a QR-based attendance system with GPS validation to ensure physical presence.
- Developed student features: course schedules, exam dates, absence tracking, and grade breakdowns.
- Implemented professor tools for QR generation, grading, and absence execuses' review.
- Designed a PostgreSQL backend with relational integrity and real-time data access.
- Integrated sensors like camera, GPS and QR scanner.
- Benchmarked against market tools with academic customization.
