Elisa — Knowledge Agent

Elisa is the team's course content specialist. Got a question about a specific lesson, a concept from a video, or something from the course documents? She'll find it, cite the source, and give you a clear answer.

Role

Elisa handles all questions related to course content and knowledge base materials. She doesn't just guess — she searches through the actual source materials first and builds her answer from what she finds. If a topic was covered in lesson 7 of module 3, she'll tell you that.

Personality

Elisa is friendly and structured. She keeps her answers well-organized and casually mentions where things come from — "This was covered in lesson 4 of the Sales module, around the 12-minute mark." She's helpful without being overwhelming.

How it works

When Elisa receives a question, she follows a structured pipeline to find the best answer:

Step 1: Generate embedding

The question is converted into a vector embedding using OpenAI's embedding model. This captures the semantic meaning of what you're asking, not just keywords.

Step 2: Unified search

The embedding is used to perform a similarity search across all knowledge sources simultaneously using pgvector. This includes:

• Video transcripts — with timestamps, so Elisa can point you to the exact moment in a video
• Documents — PDFs, guides, and other written course materials
• Resources — links, prompts, templates, and reference texts

The search returns the most relevant chunks from across all sources, ranked by similarity.

Step 3: Generate answer

The relevant context chunks and your original question are sent to Claude, which generates a comprehensive answer grounded in the actual source material.

Step 4: Clarification (if needed)

If your question is too vague or could mean multiple things, Elisa will ask for clarification instead of guessing.

Sources

Elisa searches across three types of content:

Video transcripts

Full transcripts of all course videos, broken into chunks with timestamps. When Elisa references a video, she'll include which lesson it's from and roughly where in the video the topic is discussed.

Documents

Written course materials like PDFs, worksheets, and guides. These are chunked and indexed the same way as video transcripts.

Resources

Supplementary materials including links, prompts, templates, and reference texts. These are often shorter pieces of content that support the main lessons.

Example interaction

curl -X POST https://api.dealdome.eu/query \
  -H "Authorization: Bearer {token}" \
  -H "Content-Type: application/json" \
  -d '{"message": "How should I structure a product launch email sequence?"}'

{
  "response": "Great question! According to Elisa, this is covered in detail in the Email Marketing module, lesson 3 (around 8:30). The recommended structure is: 1) Teaser email 3 days before launch, 2) Launch day announcement with the main offer, 3) Social proof email on day 2 with early results, 4) FAQ / objection handling on day 3, and 5) Final reminder with urgency on the last day. The key takeaway from the lesson is to focus each email on one single idea — don't try to cram everything into one message.",
  "agent": "alfred",
  "delegated_to": "elisa",
  "sources": [
    {
      "type": "video_transcript",
      "module": "Email Marketing",
      "lesson": 3,
      "timestamp": "8:30"
    },
    {
      "type": "document",
      "title": "Launch Sequence Template",
      "module": "Email Marketing"
    }
  ]
}

When Elisa asks for clarification

If a question is too broad, Elisa won't just wing it. For example:

You: "How does that thing with the emails work?"

Elisa: "Could you be a bit more specific? Are you asking about the email sequence structure, the technical setup with your email provider, or the copywriting templates from the course?"

This way you get a focused, useful answer instead of a vague overview.