We envision a future where artificial intelligence respects boundaries not only in its capabilities but in its memory as well, enabling a new era of trust and safety in AI deployment. By pioneering the world's first Machine Unlearning Platform, we empower enterprises to precisely sculpt and refine what their AI models retain and what they consciously forget, ensuring the integrity and reliability of AI decision-making processes.
Our mission is to transform AI into a responsible and compliant technology that actively mitigates risk by erasing biases, vulnerabilities, and inaccuracies embedded within its algorithms. Leveraging cutting-edge techniques across generative and non-generative AI modalities, we provide fast and effective solutions that enhance model performance without the costly need for full retraining.
At Hirundo, we are building the foundational tools for the next generation of AI—one that is transparent, accountable, and aligned with human values—so that organizations across the globe can confidently innovate while maintaining ethical standards and regulatory compliance in their AI-driven operations.
Our Review
We'll be honest—when we first heard about Hirundo's "machine unlearning" platform, it sounded like sci-fi. Teaching AI to forget? But after digging into what this 2023 startup actually does, we're genuinely impressed by how they're tackling one of AI's messiest problems.
Think about it: every AI model learns from massive datasets, and sometimes that data includes biases, errors, or stuff you really don't want your AI remembering. Traditionally, fixing this meant starting over with a completely retrained model—expensive, time-consuming, and frankly, a nightmare for enterprises.
The "Forgetting" Revolution
Hirundo's approach is refreshingly practical. Their platform can selectively remove problematic data, biases, or behaviors from AI models in under an hour. No full retraining required. We've seen them demonstrate a 76% reduction in bias for models like DeepSeek-R1 without hurting performance—that's the kind of result that makes CTOs take notice.
What caught our attention is how broad their solution is. This isn't just for chatbots or language models. They're handling computer vision, radar, LiDAR, and more. That versatility suggests they've built something fundamental, not just a narrow fix.
Serious Credentials Behind the Vision
The founding team gives us confidence this isn't just hype. Ben Luria's a Rhodes Scholar and serial entrepreneur, while Emeritus Professor Oded Shmueli brings decades of AI research from Israel's Technion. When academics of this caliber leave prestigious positions to build a startup, they usually know something the rest of us don't.
Their recent $8 million seed round, led by Maverick Ventures Israel, suggests investors agree. That's solid funding for a concept that was purely theoretical just a few years ago.
Who Really Needs This
We see Hirundo's sweet spot in high-stakes industries where AI mistakes aren't just embarrassing—they're dangerous or expensive. Finance, healthcare, defense—anywhere bias or hallucinations could cause real harm. The fact that multinational corporations and government agencies are already piloting their tech tells us the problem is real and urgent.
For smaller companies or less critical applications, the value proposition might be harder to justify. But for enterprises where AI trustworthiness is make-or-break? Hirundo looks like they're solving tomorrow's compliance headaches today.
Features
Machine Unlearning Platform to remove unwanted learned data, biases, and behaviors from AI models
Automated dataset optimization and quality assurance
Fast remediation fixing models in less than an hour without full retraining
Supports multiple AI modalities including generative AI, computer vision, radar, LiDAR, time series, speech-to-text, and NLP
Seamless integration via API or SaaS/VPC/on-prem on SOC-2 certified infrastructure
FAQs
Ploy was founded in 2024.
The first Machine Unlearning Platform: remove unwanted data & behaviors from your AI models
LangChain
Playpower Labs
Conversion
Soundful
neuroflash
Relume
Pixpic
