Loading case study
The route is preparing its narrative, media, and interaction state.
Owned product
Synapse connects playback, transcript moments, and source evidence in one workspace. The current NASA demo is deterministic; hosted transcription, OCR, embeddings, and durable jobs remain roadmap work.
02 · The problem
Long-form video, screen recordings, and research archives contain useful knowledge, but the evidence is hard to search, cite, review, and reuse. The product needed a path from uploaded media to a trusted, timestamped learning workspace.
03 · My role
Owned product architecture, Next.js app surfaces, typed demo data, playback-sync helpers, NASA ingest tests, and the worker-result contract/importer.
Designed the product around workspaces, media items, transcript chunks, citations, learning links, review queues, and processing jobs.
Built the deterministic NASA demo, private-library concepts, upload dashboard, invite flow, and reviewer interface.
Defined the worker JSON contract so Python processing can feed typed web experiences without manual translation.
Framed the next search layer around hybrid semantic search, OCR-derived visual notes, and reviewable timestamped evidence.
04 · Key decisions
Each decision connects an interface behavior to a system or operating constraint.
Transcript moments, source details, and playback position stay connected so a result can be inspected in context.
Official captions, authored representative notes, and deterministic matching keep the current experience testable while hosted transcription, OCR, embeddings, and durable jobs remain roadmap work.
Approve, correct, and send-back states model how future generated links and visual evidence can be reviewed before becoming trusted output.
05 · Product experience
Real product surfaces and bounded interaction studies make the workflow visible without exposing private source or pretending roadmap work is live.
Transcript context
Timeline ready at the first transcript moment.
06 · System
The current demo is deterministic: official captions, authored representative visual notes, and explicit matching rules keep its behavior testable.
The contract and reviewer states show where future generated evidence can enter without presenting planned hosted extraction as implemented.
07 · Verification and boundaries
Vitest covers typed demo data and playback-sync helpers. Python tests cover caption ingest, topic linking, result structure, and learning-link import behavior. The demo does not imply that hosted AI extraction is already running.
The MVP workshop docs define synchronized playback, transcript evidence, citations, and supporting evidence, with semantic search explicitly planned; the current demo uses deterministic substring matching.
Open system boundaryThe local Python worker path imports precomputed structured transcript chunks, citations, links, notes, assets, and job status; it is not yet a hosted extraction pipeline.
Open reviewer captureThe schema and reviewer states are designed so future AI output is not authoritative by default; current candidate links and visual notes are deterministic or representative.
Run evidence-sync demoNASA media is used for the public demo because it supports a safer public-facing evidence workspace than private customer content.
Open workspace capture08 · Result and current state
Synapse is under development with a credible, testable product shape and a typed processing boundary. It does not yet include hosted transcription, OCR, semantic embeddings, or durable processing jobs.
Continue the story
Cross-surface product ownership, data-pipeline work, and cost-conscious system judgment.