Sweet House Bakery - MVC PHP Framework Using AI to predict statistic

• Designed and launched a production-grade e-commerce platform integrating AI-driven personalization and conversational commerce.
• Built “Jenifer” Virtual Cashier — a bilingual (English/Vietnamese) chatbot powered by Google Gemini LLM and Thompson Sampling reinforcement learning for real-time upsell optimization.
• Engineered a multi-layer system:
  • Frontend: custom floating chat widget (chat.html) with voice toggle, live suggestions, and localized greetings.
  • Middleware (PHP): secure /ajax/chatask, /chatgreet, /chatreward bridges for AI interaction.
  • Backend (Python/FastAPI): microservice handling chat sessions, LLM calls, and reward feedback loops.
• Implemented a bandit learning engine (thompson.py, db.py) to dynamically adjust product recommendations based on live user behavior.
• Added context-aware greetings, multilingual persona prompts, and robust fallback logic to ensure uptime and consistent conversational tone.
• Designed and normalized AI learning tables (sessions, messages, rewards, arms) in MySQL for adaptive analytics.
• Created a future-ready architecture supporting voice interaction, personalized memory, and cost-optimized AI compute tiers.

AI Resume Parser

Custom resume parsing and scoring engine built with my own Python/scikit-learn NLP models, not a third-party LLM. It extracts structured candidate data, skills, experience, education, and match signals from PDF, DOCX, and TXT resumes, then serves validated JSON through a Flask API for database storage, ranking, recruiter review, and analytics.

AI Interview Chatbot

Standalone demo of my JobPortal interview scheduling AI, powered by local Python/scikit-learn models trained for intent detection, next-action routing, day and time preferences, and toxicity safety checks. The production engine uses Flask, Socket.IO, MySQL persistence, PIN verification, rule fallbacks, and state-specific booking modules so the chatbot can collect candidate interest, email, availability, slot selection, and confirmation while defending against abusive or unsafe messages. This demo isolates the trained model behavior without requiring a real candidate record or database connection.

Here's the system overview of how the AI interview chatbot is designed, including the PHP application, Python AI engine, trained models, database state, safety checks, and scheduling workflow.