deploy: sync code from GitHub main

deploy: sync code from GitHub main

deploy: sync code from GitHub main

deploy: sync code from GitHub main

deploy: sync code from GitHub main

deploy: sync code from GitHub main

deploy: sync code from GitHub main

deploy: sync code from GitHub main

perf: preload models at build time + fix DAPT model name

fix: DAPT model name + remove symlink deprecation

fix: add HF Space frontmatter to README

deploy: add HF model download at startup

deploy: add 3-model ensemble (DistilBERT + RoBERTa + DeBERTa)

ci: inject HF Space frontmatter

fix(ci): remove legacy LFS model breaking CI checkout

chore: retrigger CI after LFS upload

fix: bump SQLAlchemy to >=2.0.27 for Python 3.13 compatibility

style: apply ruff format to pass CI format check

fix: resolve 63 ruff lint errors failing CI

feat: Redis-backed rate limiting (Upstash) with in-memory fallback

feat(ml): 3-model ensemble (0.813 Micro-F1) + hybrid LLM tiers + safety override

docs(README): fix cross-repo link (GitHub can't follow ../README)

chore: final-audit fixes + CHANGELOG

docs(README): rewrite + extract HF Space frontmatter out of GitHub view

test: add pytest suite + wire into CI

style: ruff format the indexes migration file

perf(dashboard): collapse N+1 patterns on /stats and /patients

perf(backend): gzip compression + hot-path indexes + pool bump + alembic skip

fix: crisis alert email sends clinician to shared /results/{id}, not patient-only /screening/{id}

fix(docker): unpin pydantic so svix resolves its ModelWrapValidatorHandler import

CI: bump actions to Node 24-compatible versions

Fix CI lint + format + pinpoint pip-audit exception

H.36 — Sentry error monitoring (backend)

H.31 — Supabase Storage for profile pictures

H.32 — Alembic migrations formalization

H.38 — dependency audit in CI + CVE patches

H.34 — security headers: backend middleware + Vercel edge config

Phase E.16 — Resend webhook delivery tracking

Phase D — PDF exports (screening, patient data, clinical summary)

Phase C — smart form inputs: RxNorm + ICD-10 autocomplete, PDF upload

Phase B clinician features + system-wide tone softening

LLM prompts: lean warmth-forward while preserving substance

Crisis + safety guard: warm, gentle tone instead of cold canned responses

Phase B #3 — LLM Output Safety Guard

Fix 15+ audit-identified gaps across backend

UserProfile: expose full onboarding fields for frontend pre-fill

Add comprehensive /dashboard/patients/:id/full-profile endpoint

Add patient-facing appointment + care-plan endpoints

Add APScheduler + wire email triggers for appointments and care plans

Email: use HTTPS-hosted logo + polished template