Mozilla has announced funding for 14 AI projects as part of its Builders Accelerator program. This program launched in June and supports independent AI and machine learning engineers by offering up to $100,000 in funding, mentorship, and visibility to foster innovation in local AI where models run directly on personal devices rather than relying on cloud services.
The 2024 Cohort
From democratizing access to scientific data and enterprise-grade ML frameworks to on-device semantic search and interactive art, these projects are pioneering what it means to build AI in the open.
Application and training/fine-tuning
Projects that focus on improving and customizing AI models for better performance and real-world use.
Koel Labs – A platform that provides instant pronunciation feedback and language coaching, Koel Labs uses movie scenes to improve pronunciation in real-time.
Sartify – Designed to provide general and domain-specific AI capabilities for Swahili-speaking regions, Sartify’s Swahili-LLM supports multilingual interactions, including the Kiswa-English mix common in East Africa.
ScholasticAI – Helping users access high-quality scientific data in multiple languages, ScholasticAI uses open-source AI models to process open-licensed scientific documents. This approach ensures the accessibility of top-tier research while remaining private and reducing the need for remote servers.
Transformer Lab – Transformer Lab is a comprehensive application to build, train, and manage local LLMs on Windows, MacOS, and Linux. It simplifies working with LLMs, allowing users to train, fine-tune, evaluate, and export them across formats.
Developer tools and infrastructure enhancement
These tools and systems make software development easier and more powerful, helping developers build and manage their projects more effectively.
Ersilia Model Hub – A curated set of AI/ML models for drug discovery research in low- and middle-income countries. The models on Ersilia Model Hub can operate locally, ensuring access even in areas with poor internet connectivity and helping improve researcher access for drug discovery in the Global South.
Foyle – An AI copilot that learns from continuous feedback, Foyle uses a notebook format to simplify automation and operations tasks across workflows.
FREUD
Feature Retrieval, Editing, and Understanding for Developers (FREUD) helps developers control and understand AI model outputs by focusing on the interpretability of speech-to-text models like Whisper.
The NX Project – The NX project aims to create a scalable, distributed machine learning framework that outperforms current solutions by leveraging Elixir’s unique strengths in multitasking and fault tolerance.
Marker – Using advanced OCR for layout, text, and math detection, Marker converts PDFs to Markdown while running locally. This ensures high-quality data extraction without needing constant internet access. With companion tools Surya and Texify, Marker helps create robust training datasets and improves data processing efficiency.
Open WebUI – Open WebUI is a popular platform with over 3 million downloads. It offers an easy-to-use interface that lets users run powerful GPT-like AI models without an internet connection. The team is currently focused on building a community where users can share data and insights to improve AI models.
Theia AI IDE – Offering full transparency, customizable prompts, and the freedom to choose and host preferred language models, Theia AI IDE extends the existing Theia editor to provide a suite of copilot options and agents, enabling developers to customize their tools and workflows with on-device or cloud AI solutions.
Expressive media and data visualization
Creative projects finding new ways to display data and media, making information easier to understand and more engaging.
Ente -An on-device semantic search and image recognition that provides users with private, durable, and reliable photo storage. And Ente Photos is a fully open-source platform for securely storing family photos.
Latent Scope – Designed to embed, visualize, cluster, and categorize data using the latest advancements in LLMs. Latent Scope features a pipeline tool for processing datasets and an exploration interface for visualizing and editing categorized data. It can operate entirely locally with open-source models or integrate with popular model providers.
Tölvera – A creative tool for interactive art, Tölvera empowers artists to create and interact with dynamic, self-organizing systems. It is inspired by fields such as artificial life (ALife) and self-organizing systems. It provides creative coding-style APIs that allow users to combine and compose various built-in behaviors, such as flocking, slime mold grow.
KEY QUOTE:
“When we first opened applications, we weren’t sure how many would dive in, given that local AI is still a relatively new concept. But the open-source community proved its strength and potential beyond our expectations. These 14 projects stood out with their creativity, potential impact on the local AI tech stack, and alignment with Mozilla’s values.”
- Liv Erickson, who leads ecosystem development at Mozilla
