Docker restart

fix: update source metadata filenames from .pdf to .md format for consistency

feat: implement bibliographic citation handling with source metadata integration

fix: update default value for TOP_K and enhance prompt guidelines for balanced information presentation

feat: enhance chunking and extraction processes with debug artifact handling and function name corrections

feat: Decrease KaTeX font size and add a prompt rule requiring introductory text after headings.

feat: Update LLM prompt citation instructions to use a numbered reference list format without page numbers.

feat: Implement RAG pipeline progress visualization and enhance code block copy button feedback.

feat: add Kimi K2 and Llama 4 Scout Groq models with updated priority and quotas

feat: Improve and clarify LaTeX formatting instructions in LLM prompts for mathematical expressions.

feat: Migrate Gemini LLM integration to the new `google-genai` SDK, adopting its client-based asynchronous architecture.

Change default LLM provider from Gemini to Groq

Add Claude-style response formatting guidelines to prompt

Fix LaTeX formula rendering in chat responses

fix: Use HfApi.upload_file for more reliable qlog uploads

fix: Add better logging for qlog debugging

feat: Enable live query logging to HuggingFace

fix: Stop LLM from greeting on every response

fix: Handle appreciation messages conversationally

feat: Multiple UI and backend improvements

Fix chunk loading and huggingface-hub version issues

Add missing logger import to BM25Retriever

Fix BM25 pickle format compatibility between build and serve

Align retrieval file names with build pipeline output

Handle legacy manifest format gracefully

Fix: use dict type annotation with manual validation to avoid forward reference issues

Fix: add model_rebuild() to resolve Pydantic forward references

Fix: remove /api prefix from routes - nginx already strips it

Prepare for HF Space deployment

Initial commit for HF Space