Update README with corrected claims, add competition writeup

Update README.md

Final evaluation report and README with all corrections

Revert: remove eval from startup, restore normal operation

Remove duplicate ref_mrs_thompson with trailing space

Add LoRA evaluation on startup (RUN_LORA_EVAL flag)

Revert: remove training from startup, keep torch env fixes

Fix: remove unsupported SFTConfig params

Fix: USER env for torch cache, remove duplicate import

Fix: use SFTConfig for newer TRL API

Fix: remove deprecated dataset_text_field for newer TRL

Fix: tokenizer -> processing_class for newer TRL

Fix: set USER env var for PyTorch cache in container

Upgrade torch>=2.5 for bitsandbytes compatibility

Fix: total_mem -> total_memory in training script

Add optional LoRA training on startup (RUN_LORA_TRAINING flag)

Clarify Plan vs Advice to patient separation in prompt

Update letter format: separate Assessment/Plan/Advice, GP-reviewed structure, BLEU-optimized references

Fix: Processing Audio stays until Gradio confirms Audio Ready

Fix: stop timer on recording stop so JS Processing/Ready states persist

Task 4: Recording->Processing->Audio Ready flow with button states

Remove broken tooltip JS to restore gradient

Fix: escape nested quotes in onload JS, restore gradient

Fix: join onload JS to single line, restore gradient

Fix: restore gradient, fix clear button aria-label

Style X as re-record button with tooltip, hide download/share

Hide X/download/share/trim icons, keep re-record

Fix: instructions on separate lines, black text

Persistent Audio Ready, clear instructions, CUDA recovery

Fix: capture audio on stop_recording, show ready signal, no play needed

Fix: clear audio between patients, hide icons/label, Inter font

Fix mic CSS: real Gradio selectors, hide mic-select, gold bar, Inter font

Mic UX: gold bar controls, frosty disabled state, hide no-mic label, better text

Resample mic audio to 16kHz + style mic component to match Clarke

Show microphone on recording screen with Clarke styling

Bold all section headings in PDF export

Hide Letter Header and Sign-off section headings

Add Summary+PMH to section parser + preserve blank lines

Add blank line spacing between letter sections

Add summary paragraph + PMH section to letter prompt

Fix monkey-patch: find device arg by type not position

transformers>=4.52 + monkey-patch + CTC decode + real audio

Regenerate TTS audio files for all 5 demo patients

Fix librosa numba cache dir

Google transformers commit + manual CTC decode + revision pin

Add trust_remote_code=True to MedASR pipeline

Pin MedASR to Dec 22 revision + transformers==4.47.1

Use Google-recommended transformers commit for native LASR support

Fix: force CTC argmax decode + upgrade bitsandbytes

Monkey-patch LASR feature extractor arg mismatch