feat: add LLM query correction step to fix phonetic STT mispronunciations

fix: improve handling of short conversational affirmations

fix: double max_output_tokens to support Devanagari string lengths

fix: relax LLM length constraints to prevent mid-sentence EOS truncation

perf: implement benchmark-driven optimizations

fix(llm): remove lightbulb emoji from prompt to stop TTS from dictating it

feat(llm): add conversational memory for follow-up questions

fix(llm): prevent prompt hallucination and internal thought leaking

fix(rag): remove english translation step & increase top-k chunks

feat(llm): add context-aware follow up question to prompt

feat: session chat history with inline 🔊 audio + safety exception for fraud

fix(llm): enforce strict RAG grounding

ui & llm: rearrange layout and enforce concise chat

fix: few-shot chat translation, max_output_tokens 1024, bigger textbox

feat: conversational bot overhaul — warm persona, complete answers, no truncation

fix: robust translation (3-tier fallback), dual-query RAG, pipeline safeguards

revert: restore correct model name gemini-3-flash-preview

fix: use gemini-2.0-flash, robust Hinglish translation, RAG k=5

fix: handle NoneType response.text from gemini-3-flash-preview

fix: update Gemini 3 Flash model ID to gemini-3-flash-preview

feat: upgrade to gemini-3-flash for paid API account

fix: switch to gemini-2.5-flash (2.0 series retired March 2026)

fix: migrate from deprecated google-generativeai to new google-genai SDK

fix: update Gemini model from deprecated gemini-1.5-flash to gemini-2.0-flash

Financial literacy UX: follow-up suggestions, emerald theme, Hindi-first labels

Switch LLM to Gemini Flash (free, better Hindi, no billing issues)

Switch to Mistral-7B API: clean prompts, fix UI labels, restore HF_TOKEN warning

Fix LLM prompts: strict few-shot examples, low temp, anti-hallucination post-processing

Fix: remove duplicate Mistral functions that were overriding Qwen code

Switch LLM to Qwen2.5-1.5B-Instruct (local CPU, no token needed)

Clean rewrite: whisper-small local, clean RAG+LLM with Hindi prompts

Fix: switch to Mistral-7B, use chat completions API, use LlamaIndex + CSV

Clean commit without venv