Build searchable video archives with timestamped transcripts

A video archive becomes useful when people can search what was said, not only titles and descriptions. Timestamped transcripts turn videos into documents without losing the link back to the exact moment.

This applies to:

• webinars;
• conferences;
• training libraries;
• internal talks;
• interviews;
• product demos;
• public YouTube catalogues;
• customer research recordings.

The workflow is straightforward:

Video URL
   ↓
Transcript with segments
   ↓
Searchable chunks
   ↓
Full-text or vector index
   ↓
Result with timestamp citation

Techtuel provides the first two steps through a video transcription API.

Why titles are not enough

Video metadata is thin. A title may say "Q3 product webinar", while the answer someone needs is twelve minutes in:

"The export API supports VTT and SRT."

Without a transcript, search cannot find that sentence. Without timestamps, search can find the sentence but cannot send the viewer to the right moment.

A useful video archive needs both:

• text for retrieval;
• time offsets for citation.

Transcribe the video

Submit a video URL:

curl -X POST https://api.techtuel.com/v1/transcriptions \
  -H "Authorization: Bearer $TECHTUEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "source_url": "https://example.com/videos/product-webinar"
  }'

When the job is complete, fetch JSON:

curl "https://api.techtuel.com/v1/transcriptions/job_7c1a?format=json" \
  -H "Authorization: Bearer $TECHTUEL_API_KEY"

The useful fields are:

{
  "title": "Product webinar",
  "transcript": "Welcome everyone...",
  "segments": [
    { "start_seconds": 0, "text": "Welcome everyone" },
    { "start_seconds": 8.4, "text": "Today we are covering exports" }
  ]
}

Each segment has a start time and text. There is no end_seconds field: derive the end from the next segment when you build chunks.

Turn segments into searchable chunks

Do not index every segment alone. Individual segments are often too small.

Group adjacent segments into chunks:

• 300 to 800 words for semantic search;
• shorter chunks for exact citation;
• keep start_seconds from the first segment;
• derive end_seconds from the following segment;
• keep the video id and source URL.

Example document:

{
  "video_id": "vid_123",
  "title": "Product webinar",
  "source_url": "https://example.com/videos/product-webinar",
  "start_seconds": 482.2,
  "end_seconds": 537.6,
  "content": "The export API supports VTT and SRT..."
}

Store for search

A simple PostgreSQL schema can carry the archive:

CREATE TABLE video_transcript_chunks (
  id UUID PRIMARY KEY,
  video_id TEXT NOT NULL,
  title TEXT NOT NULL,
  source_url TEXT NOT NULL,
  start_seconds NUMERIC NOT NULL,
  end_seconds NUMERIC,
  content TEXT NOT NULL,
  created_at TIMESTAMPTZ NOT NULL DEFAULT now()
);
 
CREATE INDEX video_transcript_chunks_search_idx
  ON video_transcript_chunks
  USING GIN (to_tsvector('english', content));

Add embeddings if you need semantic search or RAG. Keep full-text search even then: exact product names, acronyms and API fields often perform better lexically.

Build timestamp links

For YouTube, timestamp links are simple:

export function youtubeTimestamp(url: string, seconds: number) {
  return `${url}${url.includes('?') ? '&' : '?'}t=${Math.floor(seconds)}s`
}

For your own player, store start_seconds and seek the video programmatically when a search result is opened.

What to show in search results

A good result should show:

• video title;
• short matched excerpt;
• timestamp;
• source link;
• "open at this moment" action;
• optionally speaker or chapter if your own system adds it.

The transcript should not replace the video. It should make the video navigable.

Use cases

Searchable video archives help teams build:

• internal knowledge bases;
• customer research libraries;
• conference archives;
• e-learning search;
• support enablement;
• media monitoring;
• RAG over recorded content.

If your archive starts from YouTube, read YouTube transcript API: captions vs audio transcription. If you want the deeper RAG version, read Build a RAG pipeline from video transcripts.

Read the API documentation

Try Techtuel without a card