Role:
Founder, creator
Duration:
4 months
Skills:
AI-enablement,
0->1
product
development,
product
strategy,
customer
discovery,
MVP
prioritization,
techincal
product
management,
experimentation
Languages:
Python,
TypeScript
(React
Native),
SQL
Tools:
Figma,
Claude
Code,
VS
Code,
Xcode,
Status:
Launched
FiFi is an independent AI-enabled product I designed and built from concept through to launch leveraging a variety of AI tools, open-source AI, and GitHub repos.
Every aspect of its development — including the system architecture, research, engineering, business model, and UX design — was completed solo as an exploration of what modern AI-assisted product development makes possible.
FiFi has served as a learning experience, expanding my depth of knowledge and experience with the full product development pipeline.
Short-form videos and image carousels have become one of the most influential drivers of fashion influence.
Millions of consumers discover fashion through platforms like TikTok and Instagram, yet the path from inspiration to purchase remains fragmented. Viewers often struggle to identify specific clothing items shown, leaving a gap between discovery and retail conversion.
Social media platforms were not designed to support product identification or shopping outside of adverts. Users have a few high-friction options if they want to find the item which often result in “forgetting about it until I come across something similar another time”.
So, how can we create a social-media-to-shopping experience so seamless it turns social media into window shopping?
LTK
10M+ downloads
4.9
LTK allows influencers to create shoppable posts and videos linking directly to products across thousands of brands.
Advantages:
- Short-form video fashion discovery
- Established creator ecosystem
- Mature monetization infrastructure
Disadvantages:
- Can’t discover fashion from content on other platforms
- All content framed as adverts
ShopStyle
500k+ downloads
4.9
ShopStyle aggregates products from retailers and provides a traditional e-commerce search and browsing interface.
Advantages:
- Extensive retailer catalogue
- Familiar e-commerce UX
Disadvantages:
- Users must search based on pre-existing intent
- Doesn’t support visual search
Shoppin
Unknown downloads
4.7
Shoppin enables users to discover clothing items through reverse image search and digitally try them on using AI.
Advantages:
- Supports visual search based on inspiration images and trends
- Allows users to visualize outfits before buying
Disadvantages:
- Doesn’t support video content
- Requires manual image upload
Interviews
“Every time I realize a video is an ad, I lose interest. It feels disingenuous.”
“I follow a lot of fashion accounts but it’s so hard to figure out where things are from... especially if <the creator> doesn’t reply in the comments.”
“I get really bored of sifting through fashion websites for things I like.”
“I end up screenshotting to search for similar items but it’s finicky getting the right bit of the video.”
“If I see an item I like on Insta, I usually give up if I can’t find it quickly.”
How do people shop from social media inspiration?
Early research focused on understanding how users currently respond when they see fashion they like on social media.
Interviews and observational research revealed several common behaviors.
Scrub & screen shot
Users often scrub, pause, take screenshots, and attempt standard reverse image searches.
Comment section reliance
Many rely on comment sections to identify brands or items, yet the creator often fails to reply.
Text search
When the exact item cannot be identified, users search for similar items using vague descriptors like: “oversized jacket”.
Forget and revisit
Friction often stops users from actively searching at the time, opting instead to keep an eye out next time they’re shopping.
These behaviors confirmed that the demand already exists — users are actively trying to identify items — but the process is inefficient.
The opportunity was in removing the investigation step entirely.
Hypothesis:
The gap between social media inspiration and retail conversion could be addressed by extracting product information from content using computer vision software, triggered whenever a user shares content to the app.
A combination of reverse image search and textual enhancement could subsequently return the same or very similar products.
This creates value for:
Consumers
- Faster discovery
- Reduced search effort
- More engaging online shopping
- More authentic discovery
Retailers
- New source of high-intent e-commerce traffic
- Clear attribution from media-driven traffic
- Content creator partnership sourcing
- Conversion from previously unreachable demand
JTBD:
When I see a clothing item I like on social media, I want to buy it quickly and easily.
The two key technical challenges in developing FiFi’s architecture come from frame sampling and deduplication. With such variety in video formats - length, number of outfit changes, number of people, cut speed, movement, transitions - I needed an approach that intelligently sampled just the right amount. Too few frames and I’d miss items, too many and I’d have a harder time deduplicating. Carousels and images are, of course, simpler. Once I have the items, I also needed to detect duplicates of the same item across different lighting, body positions, and distances.
Developed within an internal tool for pipeline building, testing, and output evaluation, the symbolic and machine learning models integrate with a YOLOV8 computer vision model and a SERP API to form the heart of FiFi’s tech stack.
Intelligent frame sampling across variable video trends
Technical challenge
Accurate clothing detection from still frames
YoloV8 + Deep Fashion2
High precision, high recall deduplication of detected items
Technical challenge
Accurate multi-modal text description + reverse image search
SERP API
All together, FiFi's pipeline has five phases:
Each stage provides contextual information to the next, relying on open-source AI and logical inference to minimize compute. This avoid expensive LLM calls entirely while achieving an 86.5% F1 score against a dataset of 1000 labelled videos and carousels. This deterministic approach also improves consistency and reduces processing time.
A central improvement in user-rated pipeline output accuracy (73% to 79%), came from an output information architecture improvement informed by an AI-augmented user research flow I developed alongside FiFi. Initially, I showed every unique item from the content in chronological/ carousel order. However, by integrating person detection, I could group items by outfit to reduce the stress on the deduplication phase while also easing the user's item selection process.
The primary flow of the FiFi app relies on a modal popup after the user shares from inside the 3rd party app. This allows the user to find exactly what they’re looking for without needing to navigate elsewhere. Anything the user shares can also be revisited inside the app
The in-app experience primarily supports viewing and revisiting the user’s recent history of shared content and discovered clothing items. The user can jump straight back in to a particular item or even revisit all the detected items from a given video, carousel, or image.
The UI emulates an e-commerce store, creating a coherent and familiar experience. However, given FiFi’s partial novelty, in-app education plays a significant role - helping the user learn how to use FiFi to it’s full potential.
Revisit your searches and videos
Choose which item you want to shop for
Compare stores, prices, and variations
Compare stores, prices, and variations
