Synapse.
Find the moment. Follow the evidence.
Synapse connects video playback, transcripts, and source references so researchers can revisit a passage in context. The current development workspace uses public NASA material.
Synapse is not ready for a live demo. Semantic search and the broader media-processing system remain under development.
- Context
- Independent media-research product
- Role
- Product design, interface, playback, and processing contracts
- State
- In development · no live demo
- Current workspace
- NASA video, official captions, and source links
The context
Useful moments
buried in hours of media.
Finding a useful passage in a long video is only part of research. You also need to revisit the moment, check the surrounding context, and identify the source. I am building Synapse to connect those steps in one workspace, starting with NASA video and official captions.
Public NASA data must stay separate from private user workspaces and credentials.
Media-processing results need a consistent, inspectable structure before broader processing can be added.
Caption, media, and metadata contracts need to remain consistent across source types.
My contribution
Connecting transcripts
to playback and sources.
I design and build the research workspace, from caption search and playback synchronization to typed media data, Python processing contracts, and verification.
- 01
Designed the product around workspaces, media items, transcript chunks, citations, learning links, and processing jobs.
- 02
Built a deterministic NASA development scenario and synchronized media workspace.
- 03
Defined a typed JSON contract for bringing Python media-processing results into the web interface.
- 04
Built explicit environment and route controls for ongoing product development.
The decisions
Keep each reference
close to its source.
01Keep evidence tied to time
Transcript moments, source details, and playback position stay connected so a result can be inspected in context.
02Define the processing contract early
Typed worker results give transcription, OCR, embeddings, and durable jobs a stable path into the product as those capabilities develop.
03Make processing results traceable
The processing design keeps source links, timestamps, confidence state, and frame evidence together so results can be checked against the original media as those capabilities develop.
05 /The experience
Read the transcript.
Revisit the moment.
The development workspace pairs NASA video with its transcript and source links. It shows the current interaction model; a public demo is not yet available.
Architecture
From source media
to a research workspace.
Select a layer to see its responsibility and the components involved.
Public Input
NASA video, official captions, and source metadata form the current development corpus.
- NASA video
- Official captions
- Source metadata
Deterministic Processing
Caption ingest, explicit topic matching, and timestamp frames produce deterministic typed result JSON. A typed worker-result contract defines how Python processing output enters the product.
- Caption ingest
- Topic matching
- Typed result JSON
- Timestamp frame capture
Product Experience
The Next.js workspace connects caption search to video playback and source links, so a passage can be checked in context.
- Media library
- Playback sync
- Transcript search
- Source links
Development Controls
Environment and route controls limit access to the development workspace. Licensing and build checks verify the media and supporting assets used in the current corpus.
- Environment validation
- Route isolation
- Licensing checks
- Build verification
Official captions, explicit matching rules, and timestamp frames keep the development scenario deterministic.
The processing contract preserves a typed path from Python media work into the product interface.
Verification
Check the timing.
Preserve the source.
The current checks cover transcript parsing, playback synchronization, and the data passed between the processing worker and web interface.
63 Vitest checks covering development data, playback synchronization, route rules, environment validation, and supporting helpers.
6 Python tests covering NASA ingest parsing, topic linking, result structure, and learning-link import behavior.
Build checks verify licenses, notices, captions, generated frames, and the absence of browser source maps.
HTTP and browser checks verify the intended route behavior for the current environment.
Supporting material
Product design record
The product model connects synchronized playback, transcript evidence, citations, and supporting sources while semantic search remains under development.
Worker contract
The Python worker contract defines structured transcript chunks, citations, links, notes, assets, and job status for the Next.js product.
08 /Progress so far
In the current development workspace, a caption result leads to its moment in the NASA video and the surrounding source context. That interaction is implemented; semantic search and broader media processing remain in development, and Synapse is not ready for a live demo.
Current workspace screens, NASA timestamp frames, 63 web tests, 6 worker tests, and route checks.
Let's discuss the work