Challenges We Ran Into
Building an AI-powered spa booking system came with several technical and design hurdles. Here are some of the key challenges we faced and how we solved them:

🧠 1. Ensuring Contextual Understanding in AI Agents
Problem:
The AI sometimes lost track of context during multi-turn conversations, leading to incorrect bookings or confusion.

Solution:
We implemented a session-based memory store using Supabase and optimized prompt engineering to preserve conversation history and intent.

🗓️ 2. Real-Time Availability Conflicts
Problem:
Multiple users could book the same time slot simultaneously due to race conditions.

Solution:
We used database-level locking and transaction control in Supabase to ensure atomic booking operations and prevent overbooking.

🎙️ 3. Voice Recognition Accuracy (Speech Mode)
Problem:
Background noise and accents affected speech-to-text accuracy, causing misinterpretation of dates/times.

Solution:
We added input validation loops and fallbacks to text confirmation, giving users a chance to review and correct inputs before finalizing.

🔐 4. Secure Role-Based Access Control
Problem:
Enforcing strict separation between user and admin actions was tricky with Supabase Auth.

Solution:
We leveraged Supabase's Row Level Security (RLS) policies and JWT-based middleware to ensure only authorized users could access sensitive data or perform admin actions.