LuciferVid/Custom_VoiceBot

ci: add build optimization step

ci: add optimization workflow

feat: add dynamic model selection and user-provided API key support

style: remove all emojis from AI responses for a professional text-only output

fix: eliminate rephraser hallucinations for single-word queries like 'what'

fix: add specific clarification trigger for fragmented voice artifacts

fix: add voice noise filtering and strict context guarding to prevent conversational hallucinations

feat: migrate to local sentence-transformers embeddings to bypass all rate limits

perf: optimize batch sizes and RPM handling for Gemini embeddings

fix: resolve critical infinite loop in text chunking logic

perf: remove artificial sleep delay to maximize processing speed on 16GB RAM backend

fix: add Hugging Face YAML configuration to README

fix: correct Hugging Face Space name to Custom_VoiceBot

ci: add Hugging Face auto-sync workflow

chore: update frontend to point to new Hugging Face Space backend URL

feat: migrate backend to Hugging Face Spaces for 16GB RAM

fix: resolve OOM crash from giant JSON payload parsing

perf: maximum theoretical embedding speed (One-Shot Embedding)

fix: aggressively handle Gemini 429 quota exhaustion

fix: massive processing speedup by fixing chunk batching logic

perf: prioritize pypdfium2 for ALL files to prevent severe slowdowns

feat: increase max page limit to 300 for annual reports/textbooks

perf: double processing speed by maxing out Gemini API batch limit

fix: throttle embedding batches to respect Gemini free-tier 15 RPM quota

fix: skip pdfplumber for large files, stop reading at 80 pages early

perf: 3-4x faster processing — bigger chunks, bigger batches, fewer pages

fix: cap PDFs at 150 pages, extend polling to 10min, remove misleading refresh msg

fix: stream processing in micro-batches to prevent OOM on large books

fix: remove auto voice playback — Listen button only

fix: disable voice auto-play by default

fix: background upload processing to prevent Render timeout

feat: upgrade document loader with 4-strategy PDF fallback and new file types

Routine health check update

fix: replace streamlit audio player with completely invisible html5 audio component and fix autoplay interruption

feat: allow general chat when no documents are uploaded

fix: harden voice transcription logic and add diagnostic logging

fix: improve mic error reporting and force sidebar toggle visibility

style: aggressively hide audio player while maintaining functionality

fix: add missing json import in frontend

fix: restore sidebar toggle, audio playback, and empty-bar summary logic

feat: implement real-time streaming and enhanced intelligence status tracking

feat: hide document uploader when repository reaches 3 files

feat: implement 3-file upload limit for stability

feat: implement session persistence and enhanced document analysis for summaries

fix: disable automatic summary to prevent unexpected speech after indexing

docs: update README with session isolation and resilience features

fix: implement exponential backoff for embedding rate limits (429)

style: clean up citations and remove session ID from source names

fix: implement session pruning, path security, and resource optimization

feat: implement session isolation and multi-user support