Replace NaN grad_norm with null in trainer_state.json

Add Kaggle training + eval notebook links to README

Sign Blog.MD as Team Invicta

Add YouTube demo link + Blog.MD writeup

submission: final results, README rewrite, eval pipeline

pivot: reduce env scope to 3 representative cases (1 per difficulty)

fix:collab notebook build

fix(phase 7): auto-detect bf16 vs fp16 (T4 doesnt support bf16)

restructured the graders

feat(phase 7): training/train_grpo.ipynb — TRL+Unsloth GRPO Colab

fix(rubric): tighten match threshold from 0.50 to 0.55

feat: composable rubrics + anti-hacking + per-step rewards + curriculum

Merge branch 'main' of https://github.com/gauri0508/med-record

fix:graders fixed so it doesnt read from the ground-truth

fix: grader plays honestly without accessing ground truth

fix: clamp all API rewards to (0.01, 0.99), add requests dep

fix:graders with values 0-1

fix: graders now produce scores strictly between 0 and 1

fix: make /reset body optional for validator

Merge branch 'main' of https://github.com/gauri0508/med-record

Updated Readme

add HF Spaces metadata to README

feat:graders,tasks,tests,envs created

feat:handoff file

feat:initial commit with structure and db