Dec 16, 2024
Year One in Review
From Research Framework to Full-Scale Research Automation Platform
Marius Constantin-Dinu
Share this article
ExtensityAI: Year One in Review
As we mark the first anniversary of ExtensityAI, we're excited to share our journey from a research repository to a comprehensive research automation platform. This milestone represents not just a year of our startup journey, but a transformative period that has seen our initial vision evolve into a groundbreaking approach to scientific discovery and research automation. This recap covers our team's growth, key milestones, technological achievements, and our vision for the future of scientific discovery.
Our Origins: From Framework to Company
ExtensityAI emerged in January 2023 with a research repository called SymbolicAI—a neuro-symbolic framework designed to combine large language models as semantic parsers within a neuro-symbolic stack. This framework, originating from Marius-Constantin Dinu's Ph.D. research at Johannes Kepler University (JKU) in Linz, established the foundation for ExtensityAI. The development represented a fundamental rethinking of how we approach the integration of artificial intelligence with scientific research.
The initial framework resembled PyTorch in its frontend design but carried a more ambitious vision: enabling complex function compositions in a neuro-symbolic, differentiable framework for workflow automation and agent orchestration. This technological foundation would later evolve into our current focus on research automation. The framework's architecture was deliberately designed to bridge the gap between theoretical innovation and practical application, setting the stage for what would become a comprehensive research automation platform.
The "Love at First Commit" Moment
A pivotal moment in our journey came when Claudiu Leoveanu-Condrei, known as "Leo" to the team, discovered the framework repository while working in Timisoara, Romania. The discovery was a significant encounter with interesting ideas that revealed new possibilities for the evolution of general AI problem solvers.
Leo's reaction was immediate and decisive. Impressed by the framework's potential, he reached out to Marius on LinkedIn, initiating what would become a transformative partnership. What followed was, as we fondly call it, "love at first commit." The synchronicity of their vision was apparent from their first interaction, leading Leo to travel to Austria for an in-person meeting. Together, they spent intensive sessions coding and expanding the framework's capabilities, laying the groundwork for what ExtensityAI would become.
This natural partnership was further strengthened when Thomas Neumayer joined the team, bringing crucial operational expertise that complemented the technical savviness of Marius and Leo. The founding trio was complete, each member bringing unique strengths that would prove essential in transforming an academic framework into a full-fledged company.
Building the Dream Team
The assembly of ExtensityAI's team represents a careful curation of talent, bringing together individuals who not only excel in their respective fields but share a common vision for revolutionizing scientific research. Each team member was selected for their unique contribution to our mission of transforming how research is conducted in the digital age.
Technical Leadership
Our technical leadership team brings together some of the brightest minds in AI and software development:
Marius-Constantin Dinu: As co-founder and technical visionary from Sepp Hochreiter's Institute, Marius brings deep expertise in neuro-symbolic AI. His academic background and practical experience serve as the cornerstone of our technical direction, bridging the gap between theoretical innovation and practical implementation.
Claudiu Leoveanu-Condrei: Coming from Amazon, Claudiu brings invaluable experience in scaling complex systems. As co-founder and applied scientist, he combines deep industry expertise with a focus on ensuring that our innovations are ready for enterprise-level deployment.
Markus Hofmarcher: Our Research Director, also from Sepp Hochreiter's Institute, brings extensive experience in bioinformatics, reinforcement learning, and machine learning. His multidisciplinary background helps guide our research initiatives in directions that combine academic rigor with practical applicability.
Robin Goupil: As Technical Director, Robin's expertise in cloud infrastructure and scalable solutions ensures that our ambitious technical goals can be realized in production environments. His leadership is crucial in building the robust systems necessary for research automation at scale.
Design & Development
The visual and interactive aspects of our platform are crafted by a team of exceptional designers and developers:
Bogdan Petrescu: As Lead Designer, Bogdan has been instrumental in creating our corporate identity and website design. His work goes beyond aesthetics, ensuring that our complex technical solutions are presented in an accessible and engaging way.
Catalina Hasnas: Our UI Developer brings our designs to life in the ExtensityAI apps. Her expertise in frontend development ensures that our user interfaces are not just beautiful but highly functional and intuitive.
Ryan Gold: Leading multiple critical projects as our Machine Learning Engineer, Ryan bridges the gap between complex AI systems and practical applications, ensuring our solutions deliver real value to users.
Business & Operations
The business side of ExtensityAI is managed by a team with proven track records in scaling technology companies:
Thomas Neumayer: As co-founder and operations lead, Thomas keeps the company running smoothly by managing operations, finances, and runway calculations. His attention to detail ensures we maintain the resources needed for sustained growth.
Robin Simons: Bringing experience in raising €6M in previous ventures, Robin's expertise in investment strategies is crucial for our growth plans.
Matt Peterman: As an Investor and Advisor, Matt provides valuable industry connections that help us navigate the complex landscape of AI technology and business.
Brad Mattick: With experience at Salesforce, Slack, and ramp, Brad's advisory role in message crafting and networking helps us communicate our vision effectively to stakeholders.
Key Milestones and Achievements
2024 Timeline
Our first year has been marked by rapid growth and significant achievements:
March 14: A landmark day when we officially founded the company and secured our first customer, story.one with Hannes Steiner. This partnership demonstrated the immediate market validation of our vision.
April 3: Generated our first invoice, marking the transition from research project to revenue-generating business.
May 1: The onboarding of Bogdan Petrescu as our full-time designer marked the beginning of our focus on user experience and brand identity.
June 11: Successfully pitched and received the AWS (Austrian Wirtschaftsservice) Grant, providing crucial early-stage funding.
June 15: Ryan Gold joined full-time, expanding our machine learning capabilities.
September 16: Catalina Hasnas joined as full-time UI Frontend Developer, enhancing our user interface development capacity.
November 1: Established ExtensityAI S.R.L. subsidiary in Romania, marking our first step in international expansion.
November: The addition of Robin Simons brought expertise in investment strategies and fundraising.
Research and Publications
Our commitment to academic excellence and innovation is reflected in our publications:
January 2024: Our first paper "SymbolicAI: A framework for logic-based approaches combining generative models and solvers" was published after a year of intensive development and bechmarking, establishing a new state-of-the-art in neurosybmolic architectures in the academic community.
May 2024: Published "LLMs Can Self-Improve on Web Agent Tasks," demonstrating our advances in machine learning capabilities.
Conference Achievements:
Acceptance at CoLLAs and ICLR workshop showcased our work to the broader AI research community
Our second paper, currently under review for ICLR 2025, continues to push boundaries in AI research
US Patent: Secured protection for our innovative workflows and processes, strengthening our intellectual property portfolio.
Community Growth and Outreach
Our community engagement has been remarkable:
Built a vibrant Discord community exceeding 400 members, creating a space for collaboration and knowledge sharing
Achieved significant research and development milestones:
Over 1,000 stars on GitHub, indicating strong community interest
More than 50 forks, showing active engagement from developers
Reached 200,000+ views on X (formerly Twitter), demonstrating broad impact
Technical Innovations and Products
Scientific Intelligence Platform
Our flagship product has evolved from its humble beginnings into a comprehensive end-to-end research automation platform. The platform's key features demonstrate our commitment to transforming scientific research:
Paper Generation: Our automated academic paper creation system goes beyond simple text generation. It integrates proper references and citations, ensuring academic rigor while significantly reducing the time required for documentation.
Book Generation: The platform's end-to-end book creation capabilities include:
Sophisticated chapter structure generation
Professional dedications and author bio creation
Comprehensive content generation
Advanced organization and formatting features
Proposal Generator: Our automated proposal creation system transforms input parameters into comprehensive, well-structured proposals, streamlining the grant application process.
Software 3.0 Architecture
Our innovative "Software 3.0 Architecture" represents a fundamental advance in research automation:
Advanced simulators that enable complex scenario testing
Sophisticated ontology management systems
Intelligent system building frameworks that adapt to research needs
Seamless integration of symbolic and sub-symbolic systems
State-of-the-art automated fine-tuning pipeline
Extensive additional capabilities for research automation
Customer Projects and Applications
story.one Project
Our inaugural customer project showcases the practical application of our technology:
Intuitive story editor for content creation
Flexible configurator interface
Comprehensive end-to-end book generation
Sophisticated chapter management
Advanced automated content generation
University of Applied Sciences Project
This educational initiative demonstrates our platform's versatility:
State-of-the-art speech-to-speech system
Advanced RAG-augmented generation
Dynamic interactive dialogue system
Intelligent curriculum-based response generation
Research Focus and Innovation
Current Research Areas
Differentiable Neuro-symbolic Framework: Our framework pushes the boundaries of AI research through:
Advanced end-to-end optimization
Sophisticated integration of symbolic and sub-symbolic systems
Complex function composition capabilities
Automated multimodal and multistep reasoning systems
Advanced AI Technologies: We're advancing the field through:
Innovative reinforcement-based meta-agent training
Advanced knowledge representation based on typed ontologies
Sophisticated multi-agent orchestration
Collaboration with Industry Leaders
Our work with notable figures and organizations has expanded our impact:
Collaborated with Michael Laux from SAP, whose contributions continue to influence our direction
Engaged in meaningful interactions with J. McCalland and R. Pascanu from Google DeepMind
Held productive meetings with Gary Marcus and other industry leaders
Participated in enlightening discussions with Tim from Machine Learning Street Talk
Community Engagement and Events
Conference Participation
Our presence in the academic and industry conference circuit has been significant:
Delivered presentations at prestigious ICLR and CoLLAs conferences
Shared insights at TaxTech
Successfully sponsored and organized the AI Austria RL bootcamp
Hosted an impactful keynote session featuring DeepMind's Pascanu
Company Culture
We've built a strong, collaborative culture through:
Hosting periodic internal hackathons to spark creativity and drive innovation
Valuing independence and empowering team members to work autonomously
Encouraging on-demand team syncing over rigid stand-up meetings or unnecessary corporate formalities
Next Steps: International Expansion
US Market Entry
Our expansion into the US market has been strategic and purposeful:
Initiated comprehensive operations and sales movement
Conducted a productive two-week networking trip covering:
Silicon Valley's innovation hub
The dynamic Bay Area ecosystem
Seattle's tech community
Established valuable connections with potential investors and partners
Created a strong presence in key technology hubs
European Presence
Our European operations continue to grow:
Maintained strong headquarters in Austria
Established Romanian subsidiary (ExtensityAI S.R.L.)
Expanded team across multiple strategic locations
Vision for 2025
Research and Technology Goals
Our technical focus areas for the future include:
Further development of our advanced differentiable neuro-symbolic framework
Reinforcement learning developments
Enhanced automated reasoning systems
Advanced knowledge representation improvements
Business Objectives
Funding and Growth:
Strategic pursuit of additional grants (FFG, AWS)
Planned release of MVP in B2C market
Targeted acquisition of first three major B2B research automation clients
Enhanced university collaborations
Infrastructure Development:
Strategic establishment of US operations
Development of proprietary cloud infrastructure
Implementation of self-hosting solutions for clients
Team Development:
Continued full-time team expansion
Regular team events (summer and winter)
Enhancement of our research culture
Looking Ahead: The Future of Research Automation
ExtensityAI's vision extends beyond conventional AI development. We firmly believe that within 5–10 years, algorithmic discoveries will become the primary driver of research advancement. Our platform aims to enable revolutionary capabilities:
"One-click" vaccine development for diseases
Accelerated protein discovery for biotech applications
Innovative material and chemical combination discoveries
Cutting-edge architectural and infrastructure innovations
State-of-the-art simulation capabilities
Sophisticated automated dataset generation
End-to-End Research Pipeline
Our platform creates a seamless research process through:
Sophisticated problem statement formulation
Comprehensive literature review
Advanced hypothesis creation
Detailed experiment design
Efficient code implementation
Precise experiment execution
Thorough data collection and analysis
Insightful result interpretation and iteration
Effective communication and publication
This automated pipeline significantly reduces the research-to-production gap, accelerating scientific discovery and innovation.
Conclusion
As we conclude our first year, ExtensityAI stands at the forefront of research automation. Our journey from a research framework to a comprehensive scientific intelligence platform demonstrates our commitment to revolutionizing scientific discovery. With a strong team, growing community, and clear vision for the future, we're well-positioned to lead the transformation of research methodology in the coming years.
We're grateful to our team, investors, partners, and community members who have supported our mission. As we move into 2025, we remain committed to our goal of democratizing scientific discovery through automation, making breakthrough research accessible and efficient for organizations worldwide.
Join us in shaping the future of scientific discovery. Follow our journey on GitHub and join our growing community on Discord.
