Skip to content

Backend Release To-Do (Backend Only)

Scope locked to backend (app/, db/, tests/, docs/). admin/ is out of scope for this cycle.

Execution Mode

  • Branch baseline: main
  • SDLC enforced: Plan → Code → Review → Verify
  • Team routing:
  • Mido: architecture + planning + reasoning-stage design
  • Abdou: review gate for all backend PRs
  • Naser: backend pipeline/retrieval implementation
  • Hadarat: multilingual I/O handling support
  • Mailo: end-to-end verification + release checks

Active Backend Issues

Plan

  • [ ] #23 — [Backend][Plan] Release blueprint for robust scrape→ingest→chat pipeline

Scraping + Ingestion Pipeline

  • [ ] #24 — [Backend][Pipeline] Robust curriculum scraper with full metadata + checkpoints
  • [ ] #25 — [Backend][Data] Persist scraped curriculum metadata to Supabase (idempotent)
  • [ ] #26 — [Backend][Ingestion] Orchestrate reference_id handoff from scrape storage
  • [ ] #27 — [Backend][Ingestion] Hybrid index + English canonicalization in ingest endpoint

Chat + Retrieval

  • [ ] #28 — [Backend][Chat] Language detect/translate in + translate out
  • [ ] #29 — [Backend][Retrieval] 75/25 hybrid search with RRF + BM25 fusion
  • [ ] #30 — [Backend][Retrieval] User-profile-aware reranking (grade/subject/tier)
  • [ ] #31 — [Backend][LLM] Reasoning-model grounded synthesis + safe multilingual output

Verify / Release Readiness

  • [ ] #32 — [Backend][Verify] Observability + end-to-end release verification

Delivery Order

  1. 23 (spec + ADRs)

  2. 24 + #25 + #26 (scrape-to-ingest chain)

  3. 27 (hybrid ingestion + English canonicalization)

  4. 28 + #29 + #30 + #31 (chat quality pipeline)

  5. 32 (release verification gate)

Definition of Done for This Cycle

  • Scraper reliably captures complete curriculum metadata and persists it in Supabase.
  • Every scrape record can be ingested through reference_id handoff.
  • Ingestion produces both semantic and full-text artifacts with canonical English content.
  • Chat supports multilingual in/out with weighted hybrid retrieval and reranking.
  • End-to-end observability + verification checks are green before release.