Skip to content

presentation_script.md

Latest commit

 

History

History
63 lines (44 loc) · 3.99 KB

presentation_script.md

File metadata and controls

63 lines (44 loc) · 3.99 KB
Raw

Presentation Guide: Hyper-Context Engine

This guide provides a structured plan, visual cues, and a word-for-word script designed for a 1.5 to 2-minute demo (approximately 225 to 300 words) of the Hyper-Context Engine.

⏱️ Timeline Breakdown

Section Duration Focus Visual Action
1. Hook & Introduction 0:00 - 0:25 The Problem & The Engine's Purpose Show the new glassmorphic login screen, sign in.
2. Behavior Ingestion & Synthesis 0:25 - 0:55 Elasticsearch & Gemini 2.0 Flash Profile Synthesis Simulate a few actions (e.g. view ML, search Quantum), click "Synthesise Profile", watch the profile update in the sidebar.
3. Conversational Adaptation 0:55 - 1:35 ADK Agent & Real-time Personalization Ask a query in the chat; point out how the response style, depth, and tone map to the sidebar profile. Open the Thought Trace.
4. Conclusion 1:35 - 1:50 Tech Stack Summary & Key Takeaways Zoom out to show the full screen, conclude.

🎤 Word-for-Word Script (Target: ~270 words / 1m 45s)

🚀 Introduction (0:00 - 0:25)

Speech:

"Hi everyone. Standard LLM assistants treat every conversation as a clean slate, missing the nuance of a user's background. We built the Hyper-Context Engine to fix that. Let's sign in to our secure front-end portal as user_001."

Visual Cue:

  • Start on the glassmorphic login page. Type user_001 and demo1234, click Sign In.

⚡ Behavior & Synthesis (0:25 - 0:55)

Speech:

"Now we're in. On the left, we track real-time behavioral data. As the user reads articles, searches, or dismisses topics, these interactions are immediately routed through FastAPI and indexed in Elasticsearch. If we click Synthesise Profile, Gemini 2.0 Flash reads these raw event logs and generates a structured user profile: analyzing interests, dislikes, preferred communication style, and technical depth."

Visual Cue:

  • Click a couple of buttons under Activity Simulator (e.g. Read ML Article and Search Quantum Computing).
  • Click the orange ✨ Synthesise Profile button in the sidebar and wait a second for the sidebar fields (User Persona, Interests, Communication Style) to populate.

💬 Chat & Thought Trace (0:55 - 1:35)

Speech:

"When a user interacts with the chat agent, the Google Agentic Development Kit (ADK) queries this live profile. For example, if I ask a general question, the agent notices the user's advanced technical depth and analytical style, adapting its explanation automatically. By expanding the Thought Trace, we can see the exact tool calls the agent made to fetch context and update the profile in real-time."

Visual Cue:

  • Type a message in the chat input: "Explain how a compiler works."
  • Once the response loads, highlight the technical language used in the answer.
  • Click to expand the 🔍 Thought Trace box and scroll through the fetch_user_context tool call.

🛠️ Conclusion (1:35 - 1:50)

Speech:

"By combining Elasticsearch's quick log querying with Gemini's reasoning and the ADK's clean agent runtime, we've built a self-learning personalization loop. Thank you, and I'd love to take your questions!"

Visual Cue:

  • Hover cursor over the sidebar profile one last time, look up at the camera/screen, and stop.

💡 Pro-Tips for a Flawless Demo

  • Pre-run the services: Make sure your FastAPI backend and Streamlit frontend are running smoothly before recording or presenting.
  • Pre-populate one user: Have user_001 populated with some events so that clicking "Synthesise" works instantly. You can demonstrate clearing/switching using user_002 if time permits.
  • Keep the Thought Trace brief: Show the Thought Trace for just 2–3 seconds to prove the agent is executing real tool calls behind the scenes without getting bogged down in JSON text.
  • Slow is smooth, smooth is fast: Move your mouse deliberately and speak at an even, confident pace.