Programs
Food Intelligence Lab
AI and Machine Learning has enormous potential to accelerate innovation in sustainable proteins, but most of the field is constrained by a shared reality: models are only as good as the data behind them.
Today, much of the most valuable data needed to train impactful models is fragmented, non-standardized, or locked away in siloed projects. Datasets are difficult to compare, and promising insights often don’t translate across labs or products. The result: slower progress, duplicated efforts, and missed opportunities for impact.
In order for AI/ML to meaningfully accelerate sustainable proteins, the field needs shared infrastructure, not just better algorithms. FSI’s AI/ML for Sustainable Proteins program exists to build that foundation.
We support the development of ML-ready, field-building resources: open datasets, benchmarks, and common task frameworks that help the field measure progress, reproduce results, and build models that translate into better products and faster development cycles.
We advance this work in two complementary ways:
Grantmaking
Partnership & Technical Support
Grantmaking
We run targeted RFPs to fund open, field-building projects, especially datasets and tools that enable broader ML innovation. Past awards include open-source datasets on protein thermostability and on the mechanical (“tasting stiffness”) features of sustainable protein products.
Partnership & Technical Support
We work directly with aligned initiatives to shape data strategy, benchmarking approaches, and model-development plans. This includes support for programs such as the Sustainable Protein Action Lab and NECTAR’s AI/ML efforts, in collaboration with Stanford University and with support from the Bezos Earth Fund.
Leadership
The program is led by Bill Meyer, with guidance from an AI/ML Advisory Board spanning research, engineering, and product development.
Get Involved
We periodically publish new funding opportunities. If you’re building open datasets, benchmarks, or ML-ready measurement frameworks that could accelerate sustainable protein innovation, we’d love to hear from you. Watch for our upcoming RFPs, or reach out to explore alignment and timing.
“Shared data is the fastest way to compound progress.”
Advisory Board
-
Peter Cnudde
Director of Engineering, NVIDIA
-
Anna Thomas
Director, Machine Learning, FSI
-
Karthik Sekar
Lead Artificial Intelligence & Operational Technology Architect, UPSIDE Foods
-
Noa Weiss
AI Research Lead, Delphos Labs
-
Benjamin Shapiro, PhD
AI/ML Technical Product, GSK
-
Itsi Weinstock
Director of Philanthropic Growth, Senterra Funders
Request for Proposals
We periodically fund a small number of innovative projects to fund that can significantly accelerate the development and adoption of alternative proteins.
Explore our recent RFP: “Creating Benchmark Datasets and Common Task Frameworks for Alternative Protein Development.”
Scroll down to read our FAQ and granting policies and see the review team. If you have a relevant and promising project or research idea that needs funding, please err on the side of applying, even if the project doesn’t completely align with our current RFP.
Funded Projects
Creating a protein thermostability dataset to accelerate cultured meat and seafood development
Principal Investigator: Breanna Duffy, PhD, Director of Responsible Research & Innovation - US, New Harvest
“Growth factor stability is critical to cultured meat production. Therefore, understanding growth factor stability across a wide range of temperatures, including production, harvest, storage, and cooking, is critical to the growth of the industry. Recent advances in artificial intelligence have led to models that help predict the stability of proteins and could accelerate the optimization of growth factors for cultured meat production. However, the lack of public data on cultured meat-relevant growth factors limits its application.
This project developed an open source dataset on growth factor thermostability compatible with AI workflows using a combination of publicly available thermostability data (gathered using a custom-developed Large Language Model (LLM)-based toolset), experimentally-derived melting temperature data, and in silico data on protein features. The open source dataset will inform the optimization of growth factors, guide culture media safety evaluations, and support the development of computational workflows to accelerate CM development.”
Tasting stiffness: An open-source dataset for the mechanical features of alternative protein products
Principal Investigator: Ellen Kuhl, Ph.D., Catherine Holman Johnson Director of Stanford Bio-X, Walter B. Reinhold Professor in the School of Engineering, Stanford University
As part of this project on Tasting stiffness: An open-source dataset for the mechanical features of alternative protein products, we have developed an open-source dataset and computational framework to systematically quantify the mechanical behavior of alternative protein products. Using a combination of uniaxial tension, compression, and shear tests and biaxial tests, we characterized plant-based and animal meats and deli meats across the full three-dimensional mechanical spectrum. We custom-designed a physics-based neural network to automatically discover physics-based models for each product and extract its stiffness and texture parameters. In addition, we have completed IRB-approved sensory texture surveys with human participants to link the measured mechanical stiffness to perceived textural attributes such as softness, hardness, fibrousness, and moistness. All experimental data, testing protocols, and analysis software have been released on our open-source GitHub platform to advance transparency, reproducibility, and innovation in the study of food texture. We have successfully secured follow up funding to gradually test more products and expand this database, for example, to include fungi-based steaks.
High-Leverage AI/ML Opportunities in Alternative Proteins
Our advisory board has identified a set of high-leverage AI/ML opportunities with the potential to meaningfully accelerate progress across the alternative protein ecosystem. These opportunities span the full value chain — from reducing ingredient costs and improving scale-up feasibility, to developing shared benchmarks for taste and sustainability, to leveraging large language models for recipe development, consumer education, and balanced information access.
We are actively exploring projects aligned with these priority areas and welcome collaboration with researchers, technologists, funders, and mission-aligned partners who share our ambition to build the AI infrastructure needed to accelerate the protein transition.
Reducing the cost of raw ingredients for plant-based products using AI/ML techniques in agriculture, processing, and supply chain optimization.
The high cost of raw ingredients is a significant barrier to the widespread adoption of plant-based products. By leveraging AI/ML techniques, we seek proofs of concept that could optimize agriculture practices, streamline processing methods, and/or improve supply chain efficiency to reduce the overall cost of raw materials. Projects in this area could include using AI/ML for precision agriculture, developing predictive models for crop yields, or creating intelligent supply chain management systems that minimize waste and inefficiencies.
Developing AI/ML models to predict the cost and feasibility of scaling up alternative protein products, enabling startups to "fail fast" and minimize costly scale-up trials.
Scaling up production from bench to commercial scale is a critical challenge for many alternative protein startups. AI/ML models that accurately predict the cost and feasibility of scaling up a given product or process could help companies make informed decisions, prioritize resources, and avoid costly failures. Projects in this area could include proofs of concept for predictive models of equipment and facility costs, creating virtual scale-up simulations, or building decision support tools that integrate technical and economic feasibility analyses.
Creating a common task framework and benchmark dataset for alternative protein development, focusing on key aspects such as nutrition, taste, cost, and sustainability.
To accelerate innovation and comparative analysis in the alternative protein industry, there is a need for standardized benchmarks and datasets that enable researchers and companies to evaluate and optimize their products. A common task framework would define key metrics and evaluation criteria, such as nutritional content, sensory properties, production costs, and/or environmental impact. Projects in this area could include curating and annotating datasets of alternative protein ingredients and products, developing standardized testing protocols, or creating open-source tools for data analysis and visualization.
Leveraging large language models (LLMs) to generate appealing, nutritionally balanced, and cost-effective plant-based recipes tailored to individual preferences and constraints.
Engaging and personalized recipe suggestions can play a crucial role in promoting the adoption of plant-based diets. Large language models (LLMs) have the potential to generate novel, appetizing, and nutritionally optimized recipes based on user preferences, dietary restrictions, and available ingredients. Projects in this area could include fine-tuning LLMs on plant-based recipe databases, integrating nutritional constraints into the language generation process, or developing user-friendly interfaces for inputting preferences and displaying recipe recommendations.
Building AI-powered tools and platforms to educate and persuade consumers about the benefits of plant-based diets and alternative proteins, such as personalized nutrition planning apps or social media bots.
Effective communication and education are essential for driving consumer awareness and adoption of plant-based diets and alternative proteins. AI-powered tools and platforms can help deliver personalized, engaging, and science-based information to a wide audience. Projects in this area could include developing chatbots or virtual assistants that provide tailored nutrition advice, creating social media content generation tools that optimize for reach and engagement, or building gamified learning apps that teach users about the environmental and health benefits of plant-based eating.
Developing open-source datasets and benchmarks for customer preferences and tastes in alternative protein marketing and advertising to support more effective campaigns across the industry.
Understanding customer preferences and tastes is crucial for creating compelling marketing and advertising campaigns that drive the adoption of alternative proteins. However, many companies lack access to comprehensive and reliable data on consumer attitudes and behaviors. Developing open-source datasets and benchmarks in this area could help level the playing field and support more effective marketing strategies across the industry. Projects could include conducting large-scale surveys and choice experiments, analyzing social media sentiment and engagement data, or creating standardized metrics for evaluating the impact of different marketing approaches.
Exploring ways to ensure a balanced representation of plant-based and animal welfare perspectives in large language models (LLMs) that will increasingly mediate access to information.
As LLMs become more prevalent in mediating access to information, there is a risk that they may perpetuate biases against plant-based diets and animal welfare considerations. Ensuring a balanced representation of these perspectives in LLMs is crucial for promoting informed decision-making and avoiding the entrenchment of harmful narratives. Projects in this area could include developing techniques for detecting and mitigating biases in LLM training data, creating evaluation frameworks for assessing the fairness and inclusivity of LLM outputs, or engaging in public education and advocacy efforts to raise awareness about the importance of diverse perspectives in AI systems.
Frequently Asked Questions
-
The deadline for submission is on the RFP, as well the date we plan to notify applicants by. We’ll continue to accept applications after the deadline, but we cannot guarantee the availability of funds or the same application response time.
-
We will consider any grant, small or large. However, we do not commonly fund grants larger than $25,000 USD. You are welcome to seek co-funders and we ask that you list additional sources of funding in the application. We may also approve an application contingent on additional funding if we are unable to fully support the project or meet the funding deficit.
-
Yes, but only applications from individuals/teams we deem qualified and with the skills needed to complete the proposed research will be considered.
-
Responses to an active RFP will be given according to the time frame listed in the RFP.
-
Grant proposals are evaluated by FSI’s AI and Machine Learning Advisors, a group of highly committed and experienced professionals with broad experience in AI, ML and Alternative Proteins. We work to avoid any conflicts of interest, and conflicted advisors will be recused. Scroll down to see who comprises the advisory group.
Policies
-
We fund summer salaries for academics and hourly work for independent contractors, and will consider funding salaried work for researchers employed at non-profits. For academics that are funded on a project basis and independent contractors, we require the application submission include hourly rates and number of hours requested for each investigator involved in the project.
We do not fund equipment purchases such as computers, phones, etc. We do fund purchases of necessary data sets.
We can cover indirect/overhead costs if the grantee’s university requires it. If you are including indirect/overhead costs in your budget, please include supporting documentation stating any university required minimum for grants coming from non-profit organizations. If a university does not have a required minimum, we can cover up to a maximum of 10% of indirect costs.
-
All projects funded are for the public benefit. As such we encourage open access publication and wide distribution, and that work be licensed under a creative commons or similar license that would allow for derivative works.
Grantees are required to make all research results, methods, procedures, data, computer code and other materials accessible to the public for others to see, evaluate, and use for further R&D.
Grantees must agree to interim reporting to identify progress and any potential or actual delays in completion of the work. The nature and frequency of such reporting will be articulated on a grant by grant basis.
Any significant changes in how the funded project is conducted, including changes in team, timeline, project focus, or budget must have prior written FSI approval or further funding of the grant may be withheld, and granted funds may be required to be returned.
-
Grantees must use public repositories such as GitHub for code, Zenodo for data, Hugging Face for models and the Open Science Framework under a Creative Commons (CC0 or CC-BY) license. These materials and descriptions should include enough detail to allow replication of results.
After receiving approval of funding, we highly recommend that projects be preregistered with a full plan on an independent public registry such as GitHub or the Open Science Framework. Preregistration must include all details of the proposed work and how any analysis will be conducted. The preregistered analyses must then be referenced in the final report including if a different or additional analysis method was used, and why.
-
Awarded funds may only be used for the purposes proposed in the grant application and agreed to as part of the grant.
-
Please be sure to review “What is your policy on transparency?” and “What are the technical requirements for transparency?” above. Additionally, at the conclusion of the project, grantees must write a short report on the project that’s available to the public and hosted on the FSI website. FSI strongly encourages publication of results in peer-reviewed scientific or professional journals where possible, and publication in open access journals with no paywall is ideal.
-
By accepting an FSI grant, all grantees agree to the following:
The Grantee must comply with reasonable requests for information about research activities in a timely manner.
The Grantee must comply with all FSI policies and procedures.
The Grantee must provide the complete results (including intended and unintended learnings), and allow FSI to disseminate results in any way FSI sees fit (see “What is your policy on transparency” and “What are my responsibilities for disseminating my project results?” above)
The Grantee must keep records and account for all use of granted funds, and submit a detailed expense report following the completion of the funded project.
-
FSI reserves the right to terminate any funded research project that is not progressing in the manner agreed to in the granting application and agreement. Please be sure to review “What is your policy on transparency?” and “What are the technical requirements for transparency?” above.
Photo credits: