UPDATE: Switch landing page to use VoiceCal LinkedIn diagram

CHANGE: Remove in-person meetings, require meeting format

FEATURE: Add phone call meeting type (no Google Meet)

FEATURE: Enforce one meeting per session and improve exit flow

FIX: Prevent phone numbers from being extracted as times

REBRAND: Change ChatCal to VoiceCal in system prompt

FIX: Update OAuth redirect URI to port 8080 for local development

REBRAND: Update UI from ChatCal.ai to VoiceCal.ai

FIX: Correct time extraction for 4-digit times with AM/PM (e.g., 1120 AM)

FIX: Remove unnecessary name confirmation for implicit extraction

REVERT: Restore max_iterations to 15 (from 25)

FIX: Add implicit name extraction for statements like "Juan Valdez."

IMPROVE: Smart name parsing and increased max iterations

UX: Add back buttons to LinkedIn diagram and fix instruction visibility

FEATURE: Add LinkedIn diagram and improve chat widget UX

🎉 MAJOR FIX: ReAct-compatible prompt enables actual tool calling

FIX: Resolve event loop cleanup issues in WorkflowHandler

FIX: Proper async event loop handling for WorkflowHandler

PROMPT: Add explicit tool calling instructions

FIX: Implement proper async WorkflowHandler pattern

FIX: Use synchronous workflow iteration instead of async

RESTORE: Tool calling functionality by removing direct LLM bypass

CRITICAL FIX: Stop LLM from echoing user context in responses

FIX: Prevent STT garbage from corrupting user names

BYPASS FIX: Direct LLM call to avoid WorkflowHandler async issues

ALTERNATIVE FIX: Capture workflow response during iteration instead of .result()

CRITICAL FIX: Resolve async workflow handler result extraction

FIX: Server-side format conversion for Groq STT compatibility

FIX: Use Groq-compatible audio codecs (AAC for MP4, Opus for OGG)

FEATURE: Smart MediaRecorder format detection for Groq STT compatibility

TEST: Try opus filename extension for Groq STT compatibility

CRITICAL FIX: Use original WebM blob directly for STT transcription

FIX: Inline base64 to blob conversion for STT audio data

REVERT: Back to original working STT implementation

FIX: Correct STT audio file format from .wav to .webm

CRITICAL FIX: Add missing base64ToBlob method for STT functionality

FIX: Resolve async event loop closure during Groq LLM API requests

FIX: Add missing File, UploadFile imports for Groq STT endpoint

FEATURE: Groq STT Integration - Replace HuggingFace with Groq Whisper

FEATURE: Enhanced TTS interruption with intelligent noise filtering

UPDATE: Change default TTS voice from Fritz-PlayAI to Cheyenne-PlayAI

CRITICAL FIX: Resolve infinite loop in name extraction with international characters

FIX: Groq TTS voice parameter validation and error handling

MAJOR: Replace Kyutai TTS with Groq TTS for improved performance and reliability

CRITICAL FIX: Agent must NOT ask for confirmation when requested time is available

FIX: Agent not booking after user confirms available time slot

REMOVE: Email confirmation logic - email now provided from landing page

FEATURE: Implement TTS speech interruption for better user experience

Add comprehensive backend debugging for email session storage

Add enhanced debugging for email parameter flow investigation