Memory Detective

Inspiration

Memory Detective started from something very personal to me.

Both my grandma and grandpa-in-law have passed away. Sometimes, I find myself wishing I could go back and revisit the moments I shared with them, even just for a little while.

There are so many memories I want to hold on to. The family gatherings, the ordinary days, and those little moments that didn't seem particularly important back then but mean so much more now.

I know some of these moments were captured in photographs or videos. The problem is, they're probably scattered across different devices, old folders, and years of family photos. I might remember a particular moment, but not when it happened or where the photo ended up.

And that made me wonder:

What if I could simply describe a memory to AI, and it could help me find that moment again?

Not just search for a face or a date, but understand the little clues I still remember.

That's where the idea for Memory Detective came from.

I initially imagined creating a memorial video of my grandparents, bringing together the moments I never want to forget. But as I explored the idea, I realized that the hardest part wasn't necessarily making the video.

It was finding those precious memories in the first place.

What It Does

Memory Detective is an AI Skill for UGREEN AI NAS that helps people rediscover meaningful photos and videos from incomplete memories.

Instead of scrolling through thousands of files, someone could ask:

"Find the photos of me with my grandma during our old family gatherings. I can't remember exactly when they were taken."

The Skill is designed to search authorized media using clues such as people, places, dates, and activities, then present relevant moments for review.

Users can choose which memories matter to them, exclude sensitive content, and organize their selections into albums with suggested storylines and captions.

Although my inspiration came from remembering loved ones who have passed away, I want Memory Detective to help people celebrate memories too, whether that's a childhood friendship, a family holiday, a graduation, or a trip they wish they could experience again.

How I Built It

I began by exploring UGREEN's Uliya AI NAS ecosystem and studying its Skill Development Standard.

Rather than building a standalone application, I developed Memory Detective as a Markdown-based Skill that coordinates Uliya's supported capabilities.

Its workflow focuses on understanding a user's memory clues, searching relevant photos and videos, organizing possible matches, and letting the user review selections before preserving them.

I also prepared a synthetic benchmark containing different memory-search scenarios, missing information, duplicates, and irrelevant media to support future testing.

The initial Skill has passed Uliya's official static validator with zero errors and warnings. Testing on a physical UGREEN AI NAS is still the next step.

Challenges I Faced

One of my biggest challenges was realizing that AI doesn't necessarily understand what makes a memory special.

A blurry photograph of someone you love might mean more than a hundred perfectly taken pictures.

I didn't want the AI to simply pick the most visually appealing photos. I wanted it to help people find the moments they care about, while giving them control over the final selections.

Another challenge was working within Uliya's supported capabilities. My original ambition involved generating complete memorial videos, but the current Skill interface doesn't document an end-to-end video-rendering workflow.

So I decided to focus on solving the problem that mattered most to me: rediscovering the memories themselves.

What I Learned

This project reminded me that technology doesn't always have to be about creating something new.

Sometimes, it can be about helping us reconnect with something we already have, but have gradually lost track of.

I also learned how important it is to build AI around real human needs, respect privacy, acknowledge uncertainty, and keep people in control.

What's Next

I hope to test Memory Detective on a real UGREEN AI NAS, improve its memory-retrieval accuracy, and eventually explore creating complete memorial and travel videos.

But personally, there's one thing I would love to achieve.

I want to be able to describe a moment I shared with my grandma or grandpa-in-law, and have Memory Detective help me rediscover a photograph or video I haven't seen in years.

I know technology can't bring them back.

But perhaps it can help me revisit some of the moments I was lucky enough to share with them.

And if Memory Detective could help someone else reconnect with memories of a person they miss, that would make building this project worthwhile.

Built With

• agent-orchestration
• ai
• ai-agents
• computer-vision
• markdown
• memory-retrieval
• metadata-analysis
• multimodal-ai
• natural-language-processing
• photo-search
• semantic-search
• ugos-pro
• ugreen-ai-nas
• uliya
• uliya-skill-development-kit
• video-search