Problem and Context

Finding the right video content isn't just about searching random keywords, it’s about precision, relevance, and efficiency. Whether for content creation, research, or media production, users often struggle with:

• Inefficient video discovery – Scattered content across multiple platforms makes it hard to find what you need.

• Lack of metadata insights – Users don’t just need videos, they need context (source, duration, summary, etc.).

• Manual search fatigue – Going through multiple platforms manually is time-consuming and inefficient.

Sure, you can manually search across YouTube, TikTok, Instagram, Twitter, Facebook, and more, but why not let an AI Agent do the work for you?

🥁🥁🥁 ANOTHER DRUM ROLL PLEASE...

Introducing: Footage Finder AI Agent v1!

• Find video content instantly across multiple platforms.

• Extract metadata (URL, duration, platform, creator).

• Generate structured summaries for quick insights.

• Deliver results in a clean, organized format no more messy searches.

Project Objective

To build an AI Agent that automates video discovery and metadata extraction across multiple platforms, helping:

• Content creators find trending videos for inspiration.

• Researchers and analysts locate relevant video references.

• Media professionals access licensed or publicly available footage.

The AI Agent ensures every video search is:

• Efficient – Automated searches eliminate manual effort.

• Comprehensive – Retrieves content from diverse sources.

• Structured – Provides URLs, metadata, summaries, and search insights.

Key Findings and Insights

I tested the AI Agent on multiple platforms and queries to see if it could find relevant content faster than manual searching.

At this stage, I’m still refining benchmarking metrics, but initial tests show:

• Significantly reduced search time compared to manual browsing (by a lot!).

• Consistent, structured results with URLs, metadata, and summaries.

There’s room for improvement, but the AI Agent already saves time and enhances search accuracy!

See details below 👇

At the end of the image carousels, you’ll see a few quick validation checks I did to ensure the AI Agent wasn’t hallucinating in its responses. Aside from some minor one second discrepancy in it's duration results,  everything works as intended. 

How I Structured the Instruction for the AI Agent

The Footage Finder AI Agent is built on a structured instruction framework designed to optimize video discovery, metadata extraction, and search efficiency. The entire system was designed to:

• Ask precise questions before searching, refining the query based on user needs.

• Use platform-specific strategies to locate the most relevant video content.

• Filter and structure metadata to ensure clean, usable results.

• Generate concise summaries to help users quickly assess video relevance.

1. Information-Gathering & Query Structuring

Before executing a search, the AI Agent clarifies the user’s intent by gathering:

• Topics & Subtopics – What kind of footage is needed?

• Platform Preferences – Should it prioritize YouTube, TikTok, Reddit, etc.?

• Time Sensitivity – Does the user need recent content or historical footage?

• Content Type & Length – Short-form, long-form, professional, or user-generated?

This ensures precise targeting before searching, avoiding irrelevant results.

2. Platform-Specific Search Execution

Different video platforms handle content differently. Instead of a one-size-fits-all approach, the AI Agent applies platform specific search techniques:

• YouTube – Prioritizes high-engagement videos with metadata-rich descriptions.

• TikTok – Identifies trending hashtags and viral short-form content.

• Twitter/X – Focuses on breaking news and real-time footage.

• Reddit – Extracts videos from highly upvoted discussions in relevant subreddits.

• Instagram & Facebook – Targets Reels, IGTV, and Facebook Watch for visual content.

• Twitch & Dailymotion – Searches for livestream archives and alternative content sources.

By leveraging each platform’s strengths, the AI finds relevant, high-quality video content.

3. Metadata Extraction & Filtering

The AI Agent extracts:

• URLs – Ensuring every result has a working video link.

• Duration – Categorizing videos by length (short-form, long-form).

• Source & Platform – Providing context on where the video was found.

• Engagement Data – Prioritizing videos based on likes, shares, and views.

This structured approach ensures that users don’t just get videos, they get context.

4. Summarization & Result Formatting

Once the videos are identified, the AI Agent generates structured summaries that help users scan and decide quickly:

• What’s in the video? (Key content highlights)

• Why is it relevant? (How it matches the search criteria)

• Where was it found? (Platform & creator details)

The final results are organized for easy consumption, with filtering options by relevance, date, or length.

Limitations Faced

• The AI Agent only works with publicly available links.

Tools Used

AI Tools: ChatGPT, Claude
Microsoft 365: Copilot Studio
Techniques: Prompt Engineering

Challenges Along the Way

Building the Footage Finder AI Agent wasn’t all smooth sailing. One of the biggest frustrations?

🚨 "An external link was removed to protect your privacy." 🚨

If the AI Agent can’t return video links, then what’s the point of a Footage Finder?😅.

Why Did This Happen?

I'm still doing my research to understand it's root cause, but an initial research is already in place and a solution which is down below!

How I Solved it

To make sure the AI Agent provides working results, I built a few workarounds:

• Modified URL Formatting – Instead of pasting direct links (which sometimes got removed), the AI returns URLs like youtube[.]com/watch?v=VIDEOID, so users can easily access them.

• Platform Specific Adjustments – The AI Agent structures its response based on where it's pulling content from to ensure links aren’t blocked.

• Fallback Search Guidance – If a direct link can’t be provided, the AI gives you everything you need to find it yourself, including:

  • Platform Name

  • Video Title

  • Creator Name

  • Search Terms to locate it manually

Final Outcome

The AI Agent now provides a way to access the content, either with modified URLs or search instructions.

but to ensure it provides the video links, add a reinforcement to your prompt and that'll do it.

The Result: A Fully Functional AI Agent That Finds Video Content FAST!

The AI Agent automates video discovery, providing:

• Efficient, structured search results.

• Metadata-rich insights for better decision-making.

• Scalability for future enhancements. 

Impact & Next Steps

What I Learned

• Advanced Prompt Engineering for AI Agents.

• Microsoft 365 Ecosystem & AI Deployment.

• Building AI Search Assistants with Scalable Optimization.

Impact

• Faster, more accurate video discovery.

• Reduced search time for content creators and researchers.

• Improved metadata structuring for easy reference.

Next Steps

• Integrate with Power Automate to enhance automation and streamline workflows.

• Experiment with different prompt structures to improve accuracy, adaptability, and response quality.

and a user manual is also created to optimize AI Agent usage.