Hamza Feroze

End-to-end design, development, and deployment of production AI systems — from data preprocessing and model selection through API deployment, containerization, and monitoring. Key Projects & Achievements: Agentic AI Financial Analysis Tool for SMEs • Architected an autonomous multi-agent system using LangChain with prompt engineering and guardrails for reliable financial intelligence. • Built a Graph RAG pipeline with vector DB storage, ingesting bank transactions and financial statements using Pandas for data preprocessing, identifying trends and flagging anomalies across interconnected financial nodes. • Integrated multiple LLMs (DeepSeek, Qwen, Allam via Groq) with evaluation metrics to benchmark accuracy and select optimal models per task. AI Chatbot for Industrial Engineering & Diagnostics • Led full-stack development of a production RAG chatbot, integrating Claude and OpenAI APIs with prompt engineering and output guardrails. • Implemented Anthropic's Contextual Retrieval with cross-encoder re-ranking and ChromaDB vector database, significantly improving retrieval accuracy. • Built REST API backend with FastAPI, Celery/Redis async processing, and React frontend with WebSocket streaming. Containerized with Docker. • Benchmarked local LLMs (Qwen, Llama3, DeepSeek) via Ollama using quantitative evaluation metrics for production model selection. AMR & Vision-Based Dispatch System • Trained and deployed a YOLO model (PyTorch) for real-time detection with custom data preprocessing pipeline. • Built REST API integration for ROS2-to-factory communication, containerized with Docker for production deployment.

Completed an intensive, project-based Robotics Masterclass covering autonomous navigation, robotic manipulation, and AI-driven perception — culminating in a fully autonomous Vision-Guided Coffee Delivery system. Checkpoint Projects: ROSbot XL — PID-Controlled Navigation & Obstacle Avoidance • Programmed autonomous maze navigation using PID control and computer vision for real-time obstacle detection in Gazebo simulation. UR3e Robotic Arm — Vision-Guided Manipulation • Progressed from hardcoded joint movements to perception-based control using Point Cloud Library (PCL) for dynamic object detection and adaptive pick-and-place. Autonomous Warehouse Robot (RB-1) • Built a three-phase autonomous navigation system: from hardcoded paths to SLAM-based mapping to full autonomous navigation using the ROS Navigation Stack. TurtleBot3 — Physical Robot Assembly & Mapping • Assembled physical robot hardware with Raspberry Pi, bridged ROS1/ROS2, and deployed SLAM (Cartographer) for autonomous exploration and mapping. Capstone: Vision-Guided Coffee Delivery (UR3e) • Designed and built an end-to-end robotic automation system integrating YOLOv8 (PyTorch) for real-time perception, MoveIt for collision-free motion planning, and custom ROS nodes for the full perception-to-action pipeline. • YouTube demo: youtube.com/watch?v=kWCi1mnxssY

Enrolled at Udacity Institute of AI & Technology, a constituent member college of Woolf — a licensed Higher Education Institution in Europe. The Master of Science in Artificial Intelligence covers advanced topics in machine learning, deep learning, natural language processing, computer vision, and AI system design, providing both theoretical foundations and practical skills for building production-ready AI solutions.