Hi, I'm Mohamed Said Aly Amin

Developed a real-time multimodal AI system to assist visually impaired users in navigating and understanding their surroundings. The solution integrates a dual-model architecture combining a fine-tuned Qwen3-VL-2B for visual reasoning and Florence-2 for OCR and object detection. The system operates in three modes—scene understanding, visual question answering, and real-time text reading—providing contextual and interactive assistance. Designed for edge deployment, the system runs entirely on consumer hardware without cloud dependency, ensuring low latency, privacy, and accessibility.

Built a multi-agent AI travel planner that transforms natural-language requests into flight information, hotel suggestions, and day-by-day itineraries. Orchestrated sequential flight-research, hotel-research, itinerary-planning, and response-generation agents with LangGraph. Integrated AviationStack live-flight data and Tavily web search, including route extraction and city/country-to-IATA airport resolution. Used Groq's Llama 3.3 70B model to generate practical, budget-aware plans, while a FastAPI backend validates travel requests and PostgreSQL checkpointing preserves thread-based workflow state. Added request timeouts, invalid-JSON handling, API error handling, and clear fallbacks for unavailable flight data.

Built a multi-agent wedding-planning application that transforms structured couple preferences into tailored, research-backed wedding plans. Architected a primary LangChain planner agent that delegates venue, vendor, logistics, budget, and market-research tasks to two search-enabled subagents. Integrated Groq for plan synthesis and Tavily web search for current planning context and recommendations. Developed a responsive Streamlit interface for brief capture, style and priority selection, budget planning, cultural requirements, and API-key configuration. Added session-based saved plan history, agent-flow visibility, runtime status feedback, and Markdown/PDF exports through ReportLab.

Built a LangGraph multi-agent content platform that coordinates request routing, web research, article planning, parallel section writing, content reduction, and optional image generation. Integrated structured LLM outputs, Tavily evidence retrieval, Gemini technical visuals, and PostgreSQL checkpointing to create reliable, stateful workflows. Developed a FastAPI/Jinja web application that streams live agent progress, research, plans, and final Markdown through Server-Sent Events. Added health checks, logging, error handling, downloadable articles, and image serving for a complete end-to-end content-generation experience.

Built an AI-powered career guidance application that analyses uploaded PDF resumes and generates profile summaries, skill-gap assessments, personalised role suggestions, interview preparation topics, job-search keywords, and 30/60/90-day career roadmaps. Integrated LangChain with Groq's openai/gpt-oss-120b model for resume-aware insights and used PyMuPDF for document extraction, with a local keyword-extraction fallback when LLM responses are unavailable. Developed a Streamlit interface with resume upload, target-role input, country selection, career-roadmap tabs, and job recommendation cards. Integrated Apify-based LinkedIn and Naukri retrieval with cost-aware live-search controls and exposed the job-fetching capabilities through an MCP server for MCP-compatible clients.

Built a RAG-powered medical chatbot using Pinecone retrieval and Groq LLMs to provide structured, patient-friendly health information. Implemented UUID-and-username isolated conversation memory with LangChain message history to support contextual follow-up questions. Developed a Flask chat interface with local avatar selection, user onboarding, responsive UI design, and Markdown response rendering. Dockerized the Python 3.12 application and configured GitHub Actions deployment to Amazon ECR and EC2 with secure environment variables.

Built an educational Arabic oud platform with an interactive Flask and PyTorch chatbot for learning oud history, tuning, maqamat, and playing techniques. Implemented NLTK-based intent classification using tokenization, stemming, and bag-of-words features, with confidence-based responses and fallback handling. Enabled CPU/GPU model inference and created a responsive AJAX chat interface with browser-session history and local media. Curated domain-specific training data to support useful, context-aware educational responses.

SmartRoute AI is a full-stack platform developed to solve complex routing optimisation problems such as the Travelling Salesman Problem (TSP) and Vehicle Routing Problem (VRP). The system enables users to apply and compare multiple computational intelligence algorithms, including Genetic Algorithms, Ant Colony Optimisation, Particle Swarm Optimisation, and NSGA-II.Built using Next.js, FastAPI, PostgreSQL, and Redis, the platform supports asynchronous task orchestration and real-time optimisation workflows. It features an interactive route visualisation dashboard and a benchmarking engine that allows users to evaluate algorithm performance across different scenarios, making it a practical and scalable solution for logistics optimisation.

Designed and developed a fully automated XBRL processing engine to handle structured financial disclosures across global regulatory taxonomies, including US GAAP, IFRS, Japan FSA, India MCA, and Korea DART. The system automates extraction of financial statements such as Income statements, Other comprehensive Income,Asset, Liability and Cash Flows Statements. eliminating manual workflows and enabling near real-time data availability.

Designed and developed an end-to-end automated PDF parsing engine to extract structured financial data from annual and interim reports across 20+ global markets, including the USA, UK, Australia, Singapore, Hong Kong, India, Malaysia, UAE, and South Africa. Leveraged Vision-Language Models (VLMs), OCR, and table detection techniques to accurately extract income statements, balance sheets, and cash flow data from highly unstructured document layouts. Engineered robust processing pipelines to handle complex formats such as multi-column documents, scanned PDFs, footnote-embedded figures, and non-standard structures, ensuring consistent and reliable data extraction across diverse reporting styles.