Revert anti-bleed prompt mitigation (ineffective)

Anti-bleed + confusion-recovery guidance when no reading passage

Add LLM-judge (MRBench) socratic-quality results

Update README: bilingual model + rigorous eval (capability + behavior)

Switch to bilingual LoRA model (educrate-qwen3-bi); hide <think>; EN+ES support

Point Space to org-hosted model build-small-hackathon/educrate-qwen3-sft

remove stray pyc

Professional minimalist UI (English, no emojis); SFT model; tighter decoding + anti-hallucination prompt

Add video/post link placeholders

Add hackathon tags (track:backyard, badges) + write-up

EduCrate — Spanish Socratic tutor (Qwen3-0.6B SFT+GRPO), runs on CPU

initial commit