PFASolve

The pollution from PFAS - colloquially termed "forever chemical" - is a global problem. It is an inevitable byproduct of industry that's prevalent all around the world. The manufacturing sector has uses of PFAS in the production of textiles for waterproof clothing and stain-resistant fabrics. In firefighting, PFAS-containing aqueous film-forming foams (AFFF) have been widely used to combat fuel fires, forming a stable film over the fire to suppress flames effectively. The electronics industry utilizes PFAS in applications like semiconductors and electronic components, where their thermal and chemical stability is beneficial. Worldwide, concentrations of PFAS can range up to 5663.3 ng/L, where the limit (under the new Biden-Harris administration), is 4ng/L. In the US, one of the only countries to enforce regulations on PFAS, 60 million people are exposed to PFOA and PFOS (the most prominent forms of PFAS); therefore, in developing countries, such as India (seeing a 28.4% increase in textile industry growth), the threat of PFAS pollution is a growing concern.

But the problem exists beyond its production: the very properties which they are produced for - chemical stability - is what poses the biggest danger to human and ecological life. PFAS do not break down organically. When left unchecked, therefore, they persist in ecosystems. To humans, they are carcinogenic, lowering fertility rates and has implications on immune response and blood pressure. Animals in contact with PFAS suffer from liver damage and birth defects.

Our solution is bipartite: our geospatial neural network, and our device that consumers can implement in their homes to filter PFAS.

Our geospatial machine learning model has been designed to predict where PFAS concentration is highest across Europe, with a focus on the UK and Thames Valley region. This component solves the insufficiency of PFAS data. The model learns from data that has the coordinates of thousands of factories, military sites and airports that have PFAS producing processes. The data set also contains PFAS samples from select locations. This allows the model to determine PFAS concentration and with remarkable accuracy. We have a 3% MSE, demonstrating extremely high efficacy. The UI of this software is very simple; a user inputs a coordinate, and the model outputs a value for PFAS. 

To test the validity of the model, we took 4 random samples ourselves from across the Thames region. We sent these to a chemical analysis company. The model was able to predict the values correctly within at most 11%, and on average 6%. Predictions generally overestimated, which is preferable to under prediction, since it encourages caution.

This model means we can target areas most in need of filtration with our device, not only maximising environmental impact, but also making selling our device easy. This is because we can approach potential customers with accurate data, illustrating the impact on them more emphatically.

Our filtering device is based off significant novel research. The device is a small, very low cost, installable for use in personal settings at home on the nozzle of a tap. It stores a small amount of granular activated carbon and allows for the flow of tap water. Following optimisation, the filter can remove 82% of PFOA and 86% of PFOS. 

To achieve this success, we conducted experimentation into maximal permissible flow rate of water through the nozzle, versus granularity of (the size of individual pieces) GAC. This was achieved first through experimentation in a commercial, but scientifically robust, home setting. Then, we tested the filtration capabilities of 4 different grades of granularity. Through mathematical optimisation, we settled on the granularity 20x50.  To confirm these values computationally, we employed the chemical fluid dynamics modeller, ETDOT. The results of these experiments were concurrent within 7% of each other. 

Given these chemical parameters, we were able to design our device. At only £6.20 to print, the production costs are stunningly low. Since each device only needs to hold at most 40g of GAC, this only adds £2.43 to the cost, giving a final value of £8.63. We used the 3D printing filament PolyTerra, which is environmentally biodegrading. Therefore, our device does not contribute to microplastic pollution. 

Therefore, our solution demonstrates both a highly effective mechanism for detecting PFAS, and then combatting the pollution in a cheap and sustainable way.

Very few countries have implemented regulations on using PFAS in manufacturing, the lucky few including the US, Germany and Sweden. Unfortunately, this means that people all around the world suffer the effects of PFAS contamination. Currently, the primary barrier to effective action is knowledge: people don't know whether their tap water is at risk of forever chemical; government and health agencies don't possess adequate information with which to pass targeted .

Our beneficiaries are residents in the UK. 

The two prongs to our solution will address the issues that they face. First, we provide comprehensive information on PFAS prone areas. This information is integral in directing both legislature and limited resources to areas that need it most. Second, we create a cheap and sustainable product to easily filter PFAS. Having optimised its construction based on filtration efficacy and flow rate, it has miniscule impact on water usage in the home, but provides up to 84% cleaner water, reducing the most prominent forms of PFAS - PFOS and PFOA - to below the WHO limit.

The groups we serve aren't characterised by any particular ethnographic or social distinction. Instead, we are aiming to help normal people who are impacted by PFAS contamination. Our algorithm has picked up areas of concern across London, from wealthy Richmond in the West, to the industrial and poorer London Docklands in the East. As residents of London and the Thames Valley region, we are thus well positioned to implement the idea on a logistical level. 

Since the communities we serve are of varied socioeconomic background, our design process has emphasised accessibility. Our device only costs £8.63, which means we cater to even the worst off in London, where the lowest decile on average earns only £14,700. Since our dataset for our machine learning model considers purely geographic data, racial biases often prevalent in predictions are non-existent.

Throughout the design process, we have consulted with a variety of figures in the community and environmental space. We have worked with local GP’s (doctors) to better understand the health repercussions of PFAS pollution and the needs of the community, chemistry professors at UCL and the University of Surrey to refine our experimentation and filtration device, and environmental organisations like River Action UK to learn how best to engage locals to encourage them to adopt our technology. We therefore believe our product has been well optimised to match the diverse needs of our community.

We've tested and verified the Machine Learning model with results from our field samples, collected at 4 locations in Greater London: Richmond, Battersea, Canary Wharf and Windsor. Though we've yet to implement it to users, the high degree of accuracy achieved - 3% MSE - suggests this as a viable prototype. 

PFASolve's filtration devices have been optimised based on its filtration efficacy and impact on user experience. We have a working prototype that is in the process of receiving feedback

As highly driven teenagers, already with a strong background in research, we believe we have the capability to scale our product to something that millions in London could benefit from. However, we think our Achilles heel lies in our financial and business know-how. Although we have already demonstrated communication skills, we want to learn more about how to bring a product to market. We want to learn from business veterans first-hand, as it is otherwise an opportunity we lack. 

We believe that to connect ourselves with the wider market, we need connections to industry to help us scale our idea. With Solve's help, this process will be accelerated, allowing us to impact more people faster. Of course, additional funds will be a welcome capital injection, but ultimately Solve's support will be substantially more valuable to the long term performance of our business. 

Right now, approaches to filtering PFAS are limited due to a complete inaccessibility of data. Where other indicators are sampled millions of times annually across the Thames Valley, in the last three years, only 120 samples of PFAS concentration have been made publicly available. As such, though the harms of PFAS are relatively well understood to be extremely toxic, it is hard to relate the contamination to individuals directly as there is this lack of data. What we innovate on, is provide an accurate and comprehensive assessment of PFAS quality in a geographic region, which allows users to better understand how much PFAS is being fed through their pipes. We are the first in the our market niche to offer this service.

Equally, we innovate with our filtration device. The current market alternatives are either expensive and extremely bulky, for usage in an industrial setting, or extremely rudimentary, often just a thin sheet of pelleted activated carbon. We plot a new and exciting course, presenting a device that is low-cost, low-weight, effective, and custom designed. 

These two innovations together make our idea truly unique, and breaking into the wider market will allow us to present these components concurrently. We will do this facing very minimal viable market alternatives.

There are two direct results of our implementation that achieve our expected impact for both consumers and policymakers.

Data Generation

• We achieve this through using under-utilised geospatial data in predicting areas high in PFAS
• This allows governmental agencies and environmental organisations (River Action UK, the Rivers Trust, whom we've liased with), in the short term, to target and direct limited resources into regulation and action in areas where it's needed most - such as regulating firefighting industries in Canary Wharf. In fact, 60% of participants from our pilot interview were unaware of the pollution's existence, further highlighting the necessity for data generation.
• In the long term, the model, when trained on updated data, creates an environment in which threats are able to be immediately addressed.

Addressing the Issue

• The knowledge of threats we generate is complemented with a proprietary device that allows the threats to be addressed. We recognised the concurrent growth in demand for PFAS filtration technologies (from the Royal Society of Chemistry, for example), with the lack of cheap and effective alternatives that we've identified through interviews with local residents. 
• Therefore, we'll advertise ourself as a disruptive technology - >87% cheaper than existing alternatives, while maintaining a similar degree of efficacy. Through direct B2C sales, we'll change the way PFAS becomes consumed in households all around the UK. 
• In the short term, we'll provide peace-of-mind to users as they consume non-threatening levels of PFAS.
• There are a host of long term benefits to health and vitality as demonstrated in Impact Essay

Our major focus is to provide clean, unpolluted water in a sustainable fashion. This is through a concurrent improvement in both awareness - achieved by our industry-leading predictive technology - and in treatment, which we tackle through our efficient and cost-effive PFASolve filter. Beyond the measures of efficacy (neural network accuracy, filtration optimisation capabilities) we’ve incorporated in our R&D process, our impact metrics will measure the extent to which we reach those goals. We’ll first measure the concentration of PFAS before and  after PFASolve treatment, utilising the United States Environmental Protection Agency (USEPA)  guidelines: 4 parts per trillion. We are convinced this is possible. But further, we want to see positive effects on the life and wellbeing of users, not only in terms of health - through wellbeing surveys, but also, more generally, in terms of the impact of PFASolve filters on their life. Our testing on flow rate and our design aspires to make it easy to use, which we hope to verify. 

The second group of customers are changemakers, those who would benefit from being able to predict where PFAS levels are high and require treatment. Here, we will influence governmental organisations (UK Environment Agency) and environmental NGOs to create 5 policies for threatened regions, be that in the form of warnings, notices or distribution of PFASolve filters. The influence our technology has on tangible change will be how we measure on tertiary impacts.

Our geospatial neural network is an ensemble model between random forest and K-nearest neighbours. The intuition behind this is clear; the data we fed into the model for training was in the form of coordinates. Thus we used a number of techniques along the process for enhancing the model's efficacy. For example, we used a quad tree data structure several times in our model's design. What this does is recursively split a 2D plane to help make identifying neighbouring coordinates easier. This meant that making connections in the training data between industrial sites and sampling locations was faster. 

Within the training of the model, we employed a brief principal component analysis method to reduce the dimensionality of our data set. This was achieved through the standardisation of the data, followed by the computing of a covariance matrix that identified correlations between different data points. Finally, we computed the eigenmatrix of the matrix to produce our principal component, which are mixtures of different but correlated initial variables. We also implemented spatial cross-validation to improve the generalisability of our model.

In the model's output, we visualised data with an interactive heatmap, using Folium software with an OpenStreetMap interface, allowing users to zoom in on their homes. 

As part of our chemical experimentation, our primary implementation of technology was the usage of Liquid Chromatography Mass Spectrometry (LC/MS) machines to generate results from our sampling. This technology ultimately informed our filtration device testing.

In particular, we first extracted the primary compounds of interest (PFOS and PFOA) from an aqueous matrix via online SPE (solid phase extraction), utilising Thermo Scientific’s EQuan LC system, equipped with a Hypersil Gold aQ pre-concentration column. These compounds are then backflushed from the pre-concentration column via a gradient run and are quantified by high resolution, accurate mass (HRAM) liquid chromatography mass spectrometry (LCMS). Through this mechanism, we were able to run 9 different filtration tests, with the verification of Chemical professors at UCL and University of Surrey.

2 people: Christopher Whitfeld and Wenqi Zhao (Eton College)

8 months since the development of the prototype and 

We are only a team of 2 and so our work has always been equitable and built around our friendship. When we expand in the future however, we are keen to engage with as diverse a group of individuals as possible to ensure that we as a team are reflective of London as a whole. 

We note that our idea is built on the principle of accessibility to clean water and we will continue to maintain this principle in the future.

The glasses will initially be B2C sales (80% of revenue) and on-demand-manufacturing, priced at £9, with expansion into B2G and B2B. Within the first year, we project to sell 8000 PFASolve filters, totalling to £72000 of revenue. With a profit margin of 10%, our social impact model is financially sustainable and cost effective. We use the additional profits purely to maintain the business and for R&D.

How will we reach our customers? We'll market ourselves as a disruptive product - pioneers in optimising GAC for PFAS filtration, which is also 80% cheaper than existing products. Our CAC and churn rate is low - PFAS is a huge 

For scalability, we'll expand into B2B and B2G sales. Specifically, this advertises our predictive algorithms to policymakers for whom the technology can aid their decisions.

Though we are still at the prototype stage we have a clear idea of our business model for the future. PFASolve is an embedded social enterprise model, as our underlying desire for cleaner water is solved by our product directly. We also cater to low-income clients primarily, as it is often in poorer regions where PFAS concentration is greatest.

Our revenue will derive from our sales of our filtration device, and since we have a ~10% profit margin, we believe we maintain a healthy source of profit for future research and development. Our market is consumers and small businesses, who are naturally enfranchised to purchase our product, as the adverse impacts of PFAS are so evident and can be easily conveyed.

Our funding strategy is blended finance, utilising governmental grants to incentivise venture capitalist investment through loan guarantees, first loss capital and more. So far we have received small grants from a school competition, and we have already achieved a highly accurate prototype. Therefore, with additional funding we believe we can accelerate our idea's progress.