Clip-on.AI

Pitch

Did you know WA state alone is missing an estimated $929 million worth of fines per year, simply because of an inability to enforce the law that fines drivers who litter? And do you know that the content of litter is a hidden goldmine of information? For example, human waste is a strong indicator of public health risk, and an increase in bedding and clothing in litter composition is highly correlated with an increase in the homeless population.

Having been collecting roadside litter for almost two years, we are determined to improve city infrastructure and urban environment by transforming problems into insights, then into solutions. Our approach would also help address the homelessness crisis, which has cost California's government a staggering $25 billion in the past five years and $5.3 billion for Washington's government in the past 11 years, by offering roadside litter picking as a new type of job — the litter data curators.

For the past two years, we've collected over 120GB of litter data across five cities in the greater Seattle area, developed a 35-class litter classification dataset, built a hardware prototype with a mounted AI camera, and built machine learning prototypes to detect and classify litter. During this process, we brought seven seemingly unrelated groups of people (roadside workers and volunteers, university professors, software/hardware engineers, policy makers, lawyers, nonprofits, and even high schoolers) together and started equipping old-fashioned work (litter cleanup) with modern AI technologies. We are on track to completely reinvent the way roadside litter surveys are conducted, all while bringing communities closer and creating opportunities for new types of AI-empowered workers.

Our business model includes multiple revenue streams:

• Revenue Sharing with Government Agencies: Using our data to increase revenues through targeted enforcement, such as litter fines.
• Subscription Model for Data and Analysis: Providing governments and organizations with actionable insights on public health, economic trends, and socio-environmental dynamics.
• Software Licensing and Hardware Sales: Supplying infrastructure providers with our AI-powered devices to collect and analyze public infrastructure data, such as litter and traffic patterns.

We believe AI should work alongside people, create new jobs, and make the world a better place by tackling practical problems such as degrading social infrastructure, diminishing sense of purpose, and crumbling environments in both urban and wilderness lands.

Intro

The story began in 2019 when I became deeply concerned about the visible degradation of our environment and the lack of resources allocated to address it. For instance, highway ramps became increasingly cluttered with litter during the pandemic, yet no action was taken. Frustrated, I joined a local Adopt-a-Highway volunteer group and began cleaning up litter near my home.

While volunteering, I noticed an untapped potential in the litter itself—it held valuable data. In May 2023, I started taking photos of the litter I collected, and patterns began to emerge. For example, well-maintained areas often had more tissue paper in their litter, while less-managed areas contained abandoned clothing. It was like studying human behavior through the traces we leave behind.

By mid-2024, my friend Autumn Yuan joined as a co-founder, and together we began leveraging these insights to tackle litter control more effectively. Our goal was not just cleaner spaces but smarter, AI-driven solutions that could enhance infrastructure planning, improve environmental outcomes, and even create new types of jobs in an AI-driven era. Through this work, we aim to advance understanding of the physical world and transform how we care for our shared environment.

Work

Since May 2023, we have collected over 120GB of litter survey data across five cities in the greater Seattle area. This effort has engaged a diverse team, including data collectors, curators, machine learning engineers, and hardware prototype designers. Our outputs include litter distribution and composition analyses, as well as a 35-class litter classification dataset that is regularly updated.

This data allows us to identify areas where litter is increasing, highlighting regions in urgent need of litter fine enforcement and intervention. Additionally, by correlating our dataset with public health and demographic data, we can uncover new, previously untapped insights to inform smarter environmental and community strategies.

Here is a sample image taken from our clip-on AI device, which can be equipped to any grabber device on the market.

Here is a sample litter pickup report, consisting of 3,000 pieces of litter grouped by 35 categories. This represents the litter composition of Kirkland, WA near Highway 405.

And based on the litter count of large (>1 foot) sized litter alone, this is at least $20k worth of litter fines being missed.

The above example is from Highway 405 exits 18–20 near the city of Kirkland. The following is from exit 5, near the city of Renton.

As you can see, the litter composition is very different between the two cities, e.g. the category of dangerous items.

About

I'm Lin, the founder of Clip-on.AI and a passionate advocate for community-driven change. My journey has always been fueled by a desire to tackle the challenges that keep me up at night — like litter, mass extinctions, and the growing sense of disconnection in people's lives.

Originally trained as a mechanical engineer, I transitioned into software and machine learning over a decade ago, working at companies like Meta and Microsoft. These experiences gave me the tools to turn complex problems into actionable solutions. With Clip-on.AI, I'm combining my technical expertise and love for my community to reimagine how we approach environmental issues, starting with smarter ways to combat litter problems.

I'm Autumn, co-founder of Clip-on.AI, and I'm deeply inspired by the potential of AI to create positive change in the world around us. My mission is to leverage AI to tackle real-world challenges and empower people to take meaningful action in their communities.

With a background in Finance and Operations, I've spent my career using financial data to drive insights and operational improvements across startups and large corporations. At Clip-on.AI, I'm passionate about turning data into actionable insights that not only enhance infrastructure planning but also inspire individuals to take ownership of their neighborhoods and make a lasting impact.

Together, we believe AI can give us unparalleled abilities to see and solve problems in ways we never thought possible, enabling us all to play a part in creating cleaner, more connected communities.

Contact

• LinkedIn
• GitHub
• Email

People

Internship

Serag Sorror — Spring 2025

Intern: Serag Sorror
Organization: Clip-on.AI
Period: Spring 2025 (completed end of May)
Full Record: View on GitHub

Key Contributions

Performance Highlights

• Rapid technical growth — Adapted quickly to Clip-on.AI’s integrated hardware-software workflow
• Clear communication — Articulated design rationale and problem-solving approaches with confidence
• Strong autonomy — Delivered project milestones independently and on schedule

Updates

Building an AI That Sees the Litter We Walk Past

March 2026 — Project Update

We're building an open-source system that uses two types of cameras — a clip-on device mounted on the grabber and AI-powered glasses worn by volunteers — to automatically detect every piece of litter picked up, classify it in detail, and generate a full report so communities can see the story behind every cleanup.

The Goal: Turning Cleanup Walks into Data

Every weekend, volunteers go out with grabbers and trash bags to clean up their neighborhoods and highway green-belts. They pick up hundreds of items, 10-25 bags of litter each time, but when the walk is over, all that information disappears into a landfill. What if every piece of litter told a story?

Sample Report: Kirkland Highway 405 Cleanup — Jan, 2025
Today we collected 5 bags of litter, took 312 picture data samples along 1.8 miles of highway green-belt. The majority was food packaging (34%) and paper waste (23%), followed by beverage containers (10%). We also found 6 hazardous items including broken glasses and discarded batteries.

That's what we're building — no manual counting, no clipboards. Just walk, clean, and get the data.

How It Works: A Two-Tier AI System

Tier 1 — The Spotter (On-Device Detection)
A lightweight object detection model runs directly on the camera hardware in real time. It spots every piece of litter in the video frame and crops it out for closer analysis.

• Clip-on camera — mounted on the grabber itself, pointing at the grabber head. It captures close-up images at the moment of pickup, with the grabber jaws (~8 cm wide) serving as a built-in size reference.
• AI glasses — worn by the volunteer, capturing a first-person egocentric view of the cleanup scene as they walk and pick up items.

Tier 2 — The Classifier (Vision-Language Model)
Cropped images are sent to a more powerful AI — a vision-language model (VLM) — that determines the category, size, material, condition, whether it's hazardous, and writes a plain-English description for each piece of litter.

Building the Dataset

The images span 14 separate cleanup sessions from December 2023 through January 2025. Each image is annotated with bounding boxes and labeled with a detailed category.

Benchmarking the Classifiers

We ran a head-to-head benchmark on 241 carefully selected litter images across three leading AI services:

Gemini won on both accuracy and cost, making it our primary cloud classifier.

Performance: Where We Stand

Object Detection (Tier 1) — Our YOLO26 detection model achieves a mAP50 score of 0.905, correctly identifying and locating litter in the frame about 90% of the time.

Classification (Tier 2) — Using Google Gemini, we processed all 4,499 images with a 100% success rate at a total cost of about $4.21.

Training a Local Model — We fine-tuned a vision-language model (Qwen3-VL-8B) using QLoRA on a single consumer GPU (RTX 3080 Ti). After four iterations, our v4 model achieved an eval loss of 0.024 with no overfitting. A working local model would reduce per-image costs by 97%.

What's Next

• Deploy on edge devices — shrink the model for real-time inference on the clip-on camera and AI glasses via knowledge distillation and optimized model export
• Connect the end-to-end flow — from data collection with GPS tagging, through detection and classification, to automated report generation for communities and city councils
• Evaluate the fine-tuned model — head-to-head comparison of our v4 local model against Gemini on unseen images

Get Involved

• Cleanup organizations — want to pilot the system on your walks? We'll provide the camera setup and handle the AI.
• ML engineers — the codebase is open source with interesting problems in fine-tuning, edge deployment, and report generation.
• Community leaders — help us understand what data would be most useful for your advocacy and planning.

Reach out at contact@npotechnologies.com or find us on GitHub.

Elements

i = 0;

while (!deck.isInOrder()) {
    print 'Iteration ' + i;
    deck.shuffle();
    i++;
}

print 'It took ' + i + ' iterations to sort the deck.';