update readme

Author: prathiusharun

README.md

@@ -82,6 +82,14 @@ M --> B
 - Active recall oriented conversation flow
 - System guided reasoning instead of direct answers
 
+---
+### UI System (Claude-style Interface)
+- Centered chat layout inspired by Claude
+- Persistent sidebar (UI-level only, no DB persistence yet)
+- Dark / Light theme toggle system
+- Chat-first minimal interface design
+- Responsive input dock with action buttons
+
 ---
 
 ### Content Transformation Pipeline
@@ -95,11 +103,13 @@ M --> B
 
 ---
 
-### Backend Protection Layer
+## Backend & Auth Protection Layer
 - Backend only AI execution
 - Environment variable protection
 - Global AI kill switch
 - API route protection
+- Google OAuth authentication via NextAuth
+- Session-based user identity system
 - Request rate limiting
 - Structured backend logging
 - Production aware request handling
@@ -154,7 +164,7 @@ M --> B
 ### Learning Mode Flow
 
 - User enters raw thought or confusion
-- Request passes authentication layer 
+- Request passes Google OAuth (NextAuth Session Layer)
 - Request passes rate limiting layer (Redis)
 - Token usage is estimated before AI execution
 - AI cost budget is validated (Redis stored quota)
@@ -268,6 +278,7 @@ The pipeline only transforms user generated understanding.
 - Stateless serverless architecture
 - Protection before execution
 - Observability over assumptions
+- UI and backend are intentionally decoupled; UI state is currently ephemeral and will later migrate to persistent session storage
 
 ---
 
@@ -277,12 +288,15 @@ The system is a production-capable MVP with core safety and cost-control archite
 
 The system has production-grade UI and core backend protections in place, but is still evolving toward full production hardening in areas like observability, real-time streaming, and session persistence.
 
+
 Current limitations include:
 - Observability is limited to structured logging only; no tracing, metrics pipeline, or production monitoring dashboard is implemented yet.
 - No distributed quota coordination (single Redis instance)
 - Cost enforcement is based on token estimation and budget reservation logic, not real-time provider billing reconciliation.
 - No automated anomaly detection or abuse pattern detection system
 - Learning history exists in the system but is not yet fully leveraged into product features such as personalization, analytics, or memory-based reasoning.
+- Chat persistence layer is not yet implemented (no database-backed session storage)
+- Sidebar chat history is currently UI-level and not yet connected to backend storage
 
 ---