We want to hire an experienced full-stack developer to build a fully functional first version of our AI-powered training and evaluation platform for retail store employees. The system will train employees using videos (uploaded only by Admin), generate AI-driven quizzes from those videos, evaluate employee answers using Hindi speech input, and provide results to store owners via dashboards and WhatsApp. The platform must be smooth, reliable, scalable (100–500 users from day one), fully support WhatsApp reporting, and be optimized for Hindi audio processing. The system must reliably support 400–500 active users from day one. ⸻ 1. User Roles A. Admin • Upload and manage all training videos • Create training modules (no need to manually write questions) • Manage stores and employees • View overall analytics B. Store Owners • Cannot upload videos • Only view: • Employee performance • AI-generated comparisons (Hindi ideal vs Hindi actual answers) • Module completion • WhatsApp performance reports (must be fully working and reliable) C. Employees • Watch training videos • Attempt quizzes • Record their answers using Hindi audio • Receive feedback from AI • Track their progress ⸻ 2. Core Features A. Training Videos (Admin Only) • Secure streaming • Anti-screen recording measures (DRM or real-time watermarking) • Mobile-friendly player • Stored securely (e.g., AWS S3) ⸻ B. AI-Generated Quizzes (Fully Autonomous) The LLM must perform all quiz creation automatically: • Extract key concepts from the training video • Generate questions based on: • What was taught in the module • Hypothetical scenarios simulating customer interactions • Generate ideal answers • Generate different sets of questions for every employee to reduce cheating • Allow Admin to review/edit/manage questions if needed (optional in V1) Employee Experience • They see completely new questions each attempt • They record answers using Hindi speech • Answers are stored and processed ⸻ C. Hindi Speech Processing Pipeline 1. Employee speaks answer in Hindi (mic recording) 2. Audio is uploaded securely 3. Transcription using Whisper / Google Speech-to-Text (Hindi) 4. Transcribed text → AI comparison model 5. AI compares employee answer with its own ideal answer 6. AI generates: • Score (0–100) • Short feedback • Bullet points of missing elements Important: ✔ Evaluation is only semantic/language-based ✘ No tone analysis ✘ No hesitation detection ✘ No voice emotion scoring ⸻ D. WhatsApp Integration After each module is completed: • Store owner gets an automated WhatsApp report: • Employee’s Hindi answer • AI-generated ideal answer • Score • Missing points • Recommended improvements • Uses WhatsApp Cloud API / Twilio • WhatsApp report system must be fully functional, stable, and production-ready ⸻ E. Dashboards Store Owner Dashboard • View performance of their employees • Hindi answer vs ideal answer (side-by-side) • Scores, progress, feedback • WhatsApp logs • Completion analytics Admin Dashboard • Upload/manage videos • Module creation • System insights • Store and employee management ⸻ 3. Technical Requirements Backend • Scalable backend (Node.js / Python) • Fast Hindi audio processing • Job queues for audio/AI tasks Frontend • Clean, responsive Next.js UI • Easy recording system for Hindi speech • Attractive, minimal dashboards Cloud & Deployment • AWS (S3 + EC2/Lambda) or GCP • Video streaming + secure access • Smooth performance for 100–500 users • System must reliably support 400–500 active users from day one ⸻ 4. Additional Required Capabilities • LLM must extract content from the videos (via transcription or summarization) • LLM must generate module-specific, employee-specific questions • Questions should mix: • Memory-based • Understanding-based • Scenario-based • No two employees should get the same questions • System should prevent cheating as much as possible ⸻ 5. What We Will Provide • Training videos We will not provide questions or answers — AI must generate them. ⸻ 6. Budget Our budget is around ₹1,00,000. ⸻