Final submission: Finalized README with artifacts and video link

Final submission: Finalized README with artifacts and video link

Docs: Update theme and local run instructions

Fix: Upload local metrics.jsonl to HF space to provide pre-trained telemetry stream

Fix: add git to apt-get to allow unsloth git resolution

Fix: Lock Hugging Face ecosystem dependencies and install Unsloth natively to eliminate pip backtracking

Fix: Bypass PyTorch cu126 metadata bug with cu124

Final deployment: HF Docker fix with clean LFS history

feat(ui): implement SSE playback queue for gradual live stream simulation and fix Dockerfile CMD for HF deployment

fix(ui): enforce verbatim user terminology for charts, hook evaluate script to real dataset for dynamic threat feed, and purge last mocked strings

fix(ui): remove vestigial math dataset generation, sync evaluation script to 120 steps, and truncate metrics log on run

fix(ui): enforce strict real data telemetry, remove mocked endpoints, format charts, and log raw SSE streams

chore(training): optimize GRPO params for sub-4h target on RTX 4070

feat(security): apply phase 2 system hardening patch

feat: stable training loop and dashboard UI fix

feat: merge UI dashboard routes into core API server to support HF single port limit

fix: anchor env/ in gitignore to prevent excluding src/env package

fix: add missing __init__.py files to resolve implicit namespace package import errors in linux

fix: hardcode PYTHONPATH for Uvicorn module resolution

fix: sanitize git history, forcefully exclude outputs and libs

fix: optimize GRPO trainer, ignore checkpoints and binary libs

fix: restore space metadata

UI A/B comparison, Updated READMe file, updated RL, Need to fix errors with train_grpo.py

stable UI and major bug fixes

Update

Grand Finale Update: Dynamic RL Guardrails, Telemetry Dashboard, and Orchestrator