Ionosphere.tv
docs: add README
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+62
+62
README.md
+62
README.md
···
1
+
# Ionosphere
2
+
3
+
**A semantically enriched conference video archive for [ATmosphereConf 2026](https://atmosphereconf.org).**
4
+
5
+
[ionosphere.tv](https://ionosphere.tv)
6
+
7
+
Ionosphere turns conference recordings into a browsable, searchable, interconnected archive. Every talk is transcribed with word-level timestamps, enriched with extracted concepts and topics, and linked to the social conversation that happened around it on Bluesky.
8
+
9
+
Built as an entry for the [Streamplace](https://stream.place) VOD JAM.
10
+
11
+
## What's here
12
+
13
+
**Talks & Transcripts** — 95+ talks with synchronized video playback and word-level transcripts. Click any word to jump to that moment. Leave comments and reactions anchored to specific passages.
14
+
15
+
**Concordance Index** — A multi-column concordance of every significant term across all talks, with cross-references and timecode links. Scroll horizontally through the index and click any entry to watch the relevant moment.
16
+
17
+
**Community** — A curated overview of what people said about the conference on Bluesky: top posts, blog recaps with OG previews, conference photos, YouTube talk uploads, 20+ VOD JAM sites, and 80+ projects featured at the conference. Built by pulling ~2,000 posts from the Bluesky search API, extracting links and images, matching posts to talks by speaker mentions and timestamps.
18
+
19
+
**Mentions** — On each talk page, a sidebar shows Bluesky posts that mentioned the speaker during their talk, time-aligned with the transcript. Thread replies expand inline. Click a mention to seek the video to that moment.
20
+
21
+
**Speakers & Concepts** — Browse talks by speaker or by extracted concept. The NLP pipeline identifies named entities, topics, and relationships across talks, linking them into a navigable web.
22
+
23
+
## Architecture
24
+
25
+
Ionosphere is built on [AT Protocol](https://atproto.com). Talks, speakers, transcripts, and concepts are published as AT Protocol records on a PDS, making the entire archive addressable and interoperable with the atmosphere ecosystem.
26
+
27
+
- **Appview** — A Hono server that indexes records from Jetstream, serves XRPC endpoints, and manages a SQLite database. Enrichment pipelines handle transcription (Whisper), sentence/paragraph segmentation, named entity recognition, concept extraction, and speaker diarization.
28
+
- **Frontend** — A Next.js app with synchronized video playback (HLS.js), scroll-synced transcript rendering, OAuth-based commenting, and multi-column layouts using greedy column-fill algorithms.
29
+
- **Data** — Conference schedule ingested from the ATmosphereConf calendar, VOD recordings from Streamplace, transcripts aligned to talk boundaries using diarization-based chunking.
30
+
31
+
## Development
32
+
33
+
```bash
34
+
# Start local PDS + Jetstream
35
+
cd apps/ionosphere-appview
36
+
docker compose up -d
37
+
38
+
# Start appview
39
+
PORT=3001 npx tsx src/appview.ts
40
+
41
+
# Start frontend (use 127.0.0.1 for OAuth)
42
+
cd apps/ionosphere
43
+
NEXT_PUBLIC_API_URL=http://localhost:3001 npx next dev
44
+
```
45
+
46
+
## Deployment
47
+
48
+
Deployed on [Fly.io](https://fly.io) with persistent SQLite storage.
49
+
50
+
```bash
51
+
flyctl deploy --config fly.appview.toml --remote-only
52
+
flyctl deploy --config fly.web.toml --remote-only
53
+
curl -X POST https://api.ionosphere.tv/xrpc/tv.ionosphere.invalidate
54
+
```
55
+
56
+
## Author
57
+
58
+
Built by [Blaine Cook](https://bsky.app/profile/blaine.bsky.social).
59
+
60
+
## License
61
+
62
+
MIT