Session profile
- Audience: technical leads and trainees operating reconstruction pipelines.
- Duration: 80 minutes lecture + 10 minutes architecture critique.
- Output: pipeline architecture diagram with release and rollback policy.
Slide-by-slide lecture plan
- Slide 1 (2 min): Title and architecture mindset
- Slide 2 (5 min): Why connectomics is a systems problem
- Slide 3 (6 min): Reference architecture overview
- Ingest, transform, inference, post-process, serving.
- Slide 4 (6 min): Data contracts between stages
- Slide 5 (6 min): Orchestration and idempotency
- Slide 6 (6 min): Provenance schema design
- Slide 7 (6 min): Storage and chunking strategies
- Slide 8 (6 min): API/query layer for proofreading and analysis
- Slide 9 (6 min): Release engineering for model updates
- Slide 10 (6 min): SLOs and monitoring dashboards
- Slide 11 (5 min): Failure modes
- Non-determinism, provenance drift, hotspot bottlenecks.
- Slide 12 (5 min): Cost-performance tradeoff framework
- Slide 13 (5 min): Case study: bad release and rollback plan
- Slide 14 (6 min): Bridge to biological interpretation units (05-07).
- Primary shortlist:
course/units/figures/04-volume-reconstruction-infrastructure-selected-v1.md.
- Include one pipeline diagram and one QC dashboard mockup.
Speaker notes (expert-level)
- Treat reproducibility as a first-class feature, not documentation afterthought.
- Show how region-scoped reprocessing prevents full-pipeline reruns.
Assessment and artifacts
- Deliverable: staged architecture plan with lineage fields.
- Rubric dimensions: reliability, traceability, scalability, and recovery readiness.
Connections
Slide source file
- Marp draft source:
course/decks/marp/04-volume-reconstruction-infrastructure.marp.md
- Batch render helper:
./scripts/render_marp.sh
