Fix: add clean topic names for LLM while preserving rich keyword search variants

Refactor intent map with human-friendly topic names to fix short query ambiguity

Fix: rewrite short/vague queries for LLM to prevent false no-info responses

Enhance sandwich deduction intent mapping and query expansion logic

Update app/admin/templates/admin.html

Update app/admin/templates/admin.html

Add br between list items for vertical spacing

Fix list spacing: compact single-line prompt example + thorough br/whitespace cleanup inside ul/ol

Strip stray br tags after ul/ol opening tags to remove top space in lists

Use MutationObserver JS to reliably remove top spacing in chat bubbles

Broaden CSS selectors to fix top spacing in Gradio 5 chat bubbles

Remove leading space in chat bubbles via CSS margin fix and lstrip post-processing

Fix full stops, consistent Format B for leave queries, unpaid leave retrieval

Cap Format C at 4 lines, restore leave-name = count only intent

Enforce Format C for all policy questions with bold headers and line breaks

Restore structured explanation format with bold headers and line breaks

Allow 'can I take leave' intent and enforce bolding for key terms

Update bolding rules in system prompt for formal prose

Refine system prompt for better list formatting and formal tone

Fix formatting inconsistency and add support for greetings

Show actual API error message instead of generic rate limit

Migrate from Groq to OpenAI 4o-mini

Fix admin panel showing 'default name': update user_name in-place on rename, improve session list display

Eliminate first Groq API call: replace LLM query rewrite with instant local keyword expander (~3-6s saved per request)

Optimize RAG pipeline for speed: persistent HTTP client, faster reranker, and result caching

Fix reranker filtering out casual/sick leaves by enriching rerank query with expanded LLM keywords

Remove type='messages' from Chatbot for Gradio 5+ compatibility

Refine RAG pipeline for comprehensive leave retrieval, improve system prompt, and fix Gradio HTML rendering

feat: enhance RAG intelligence - intent-first protocol, /no_think for Qwen3, smart rerank fallback, context word cap

Push requested version of llm.py with monologue suppression and history optimization

Hard revert of llm.py to state 034bc73 as requested

Reverting to previous stable state after failed specificity experiment

Enforce high specificity and strict question answering in system prompt

Optimize token usage to resolve 413 Payload Too Large error

Strip internal reasoning monologues and increase token headroom

Optimize LLM conciseness and fix response cutoff

Fix Groq 413 Payload Too Large by strictly capping context and tokens

Refine for Extreme Conciseness: enforce word economy and dense summarization

Optimize for Intelligent Completeness: 10 context chunks and dense system prompt

Fix latency: switch to bge-small and bge-reranker-base for CPU optimization

Fix bot behavior: handle greetings, strip reasoning monologues, and sanitize history

Fix Groq 413 Payload Too Large by reducing max_tokens and context chunks

Update Groq Qwen model ID to qwen/qwen3-32b

Switch entirely to qwen-2.5-32b due to DeepSeek-R1 string deprecations

Update DeepSeek-R1 model alias to deepseek-r1-distill-qwen-32b

Update decommissioned DeepSeek-R1 model alias to specdec

Fix Groq 400 Bad Request for DeepSeek-R1 by adjusting temperature

Fix ui_gradio LLMService initialization and update README

Implement DeepSeek-R1, BGE-Large, and Hybrid Search

Update SYSTEM_PROMPT to use bullet points conditionally