Full-Stack + AI

Physical AI & Humanoid Robotics Textbook

Panaversity Hackathon I submission: a production-grade AI-native textbook platform teaching Physical AI and Humanoid Robotics across 13 weeks. Covers ROS 2, Gazebo, NVIDIA Isaac Sim, and a Conversational Humanoid capstone. Backend deployed on Hugging Face Spaces; FastAPI + SQLAlchemy + Qdrant Cloud with BAAI/bge-small-en-v1.5 FastEmbed for RAG; Groq LLaMA 3.3 70B for answers.

// Problem

Physical AI — AI systems that operate in the real physical world — is the next frontier after LLMs, yet no structured, accessible textbook exists for it. Existing robotics courses are scattered across dozens of resources, assume identical hardware for every learner, and offer zero in-context AI help when a student gets stuck. A student with an RTX 4080 workstation and one with a Jetson Orin Nano edge kit need entirely different guidance. Pakistani learners face an additional barrier: dense English-only material with no mechanism to get real-time clarification in their native language.

// Solution

A full-stack AI-native textbook platform with four progressive modules: (1) ROS 2 — nodes, topics, services, URDF, Python-ROS controllers; (2) Gazebo & Unity — physics simulation with LiDAR, depth cameras, IMUs; (3) NVIDIA Isaac — photorealistic sim, Isaac ROS, hardware-accelerated VSLAM, Nav2 path planning, sim-to-real RL; (4) Conversational Humanoid capstone — LLM-powered natural language robot interaction. Platform features: RAG chatbot (Qdrant + FastEmbed BAAI/bge-small-en-v1.5 + Groq), hardware-aware personalization at signup across 3 rig profiles, one-click Roman Urdu chapter translation, JWT auth with 3-step hardware onboarding, admin dashboard with Qdrant reindexing.

// Screenshots

RAG chatbot in action — highlight any text, click Ask AI, get an instant contextual answer from Groq LLaMA 3.3 70B

3-step hardware onboarding — Account → Skill Level → Hardware rig selection at signup

Hardware Dashboard — personalized learning path configured per rig: Sim Rig, Jetson, or Unitree G1

Live ROS 2 chapter — Python Node implementation with rclpy, actual code running on the platform

13-week curriculum roadmap with per-user progress tracking — 2/13 weeks completed, saves to PostgreSQL

Chapter page with hardware-aware content — Sim Rig label active, Ask AI button inline with the text

Full backend API — Authentication, Student Profile, Chat, Admin, Personalization, Translations, Curriculum endpoints

// Tech Stack

[Docusaurus v3][React 19][TypeScript][Tailwind CSS][FastAPI][Python 3.12][FastEmbed (BAAI/bge-small-en-v1.5)][Qdrant Cloud][Groq LLaMA 3.3 70B][SQLAlchemy][Alembic][PyJWT][GitHub Actions][GitHub Pages][Docker][Hugging Face Spaces]

// Metrics

✓13-week curriculum across 4 modules (ROS 2 → Gazebo → Isaac → Conversational Humanoid)
✓Qdrant Cloud collection 'physical-ai-docusaurus-textbook' — BAAI/bge-small-en-v1.5 384-dim vectors
✓9 SQLAlchemy models: User, StudentProfile, HardwareConfig, CurriculumProgress, ChatSession, ChatMessage, CurriculumWeek, IngestionLog, ReindexJob
✓3 hardware profiles with fully personalized content paths (RTX Sim Rig, Jetson Edge Kit, Unitree G1)
✓All hackathon bonus deliverables completed: Auth + Personalization + Urdu Translation

// Highlights

→RAG pipeline: FastEmbed (BAAI/bge-small-en-v1.5) embeds queries → Qdrant cosine search → Groq LLaMA 3.3 70B generates answer — highlight any text → Ask AI
→GrokClient auto-detects provider from API key prefix: gsk_ routes to Groq, xai- routes to xAI Grok — same code, zero config change
→Hardware-aware onboarding at signup: content adapts to RTX workstation, Jetson edge kit, or Unitree G1
→One-click Roman Urdu translation for every chapter — translations API endpoint backed by Groq
→Admin dashboard with curriculum ingestion pipeline and Qdrant reindexing endpoint
→Full CI/CD: GitHub Actions auto-deploys Docusaurus to GitHub Pages; backend on HF Spaces

Live Demo GitHub