A 3d box featuring Hirundo's logo, a swallow bird

Making ai forget

Remove unwanted data and behaviors from your AI models

Hirundo is the first Machine Unlearning platform, removing AI issues at their core

Our Solutions

LLM Behavior Unlearning

  • Mitigate biases
  • Remove hallucinations
  • Eliminate jailbreaks
Learn more

LLM Data Unlearning

  • Remove PII and sensitive data
  • Guaranteed removal at scale
  • No collateral damage
Learn more

Data QA

  • Detect and resolve mislabels
  • Increase model accuracy
  • Supporting vision, radar, timeseries, NLP, STT and tabular
Learn more
Trusted results

Eliminate LLM data leakages, biases and hallucinations

Unlearning is like neurosurgery for AI. What if you could edit out anything that shouldn't be there?

Book a demo
55% reduction in hallucinations

Our unlearned models are up to 55% less likely to hallucinate, verified on benchmarks like HaluEval.

85% reduction in jailbreaks

They are up to 85% more protected against prompt injections, verified on benchmarks like PurpleLlama.

70% reduction in biases

Our unlearned LLMs achieved up to 70% reduction in biases, verified on benchmarks like BBQ.

100% PII removed

We achieved 100% removal of fine-tuned PII from LLMs, with zero impact on other data or functionality.

Your AI could, and should, be better

Unlearn any pre-trained or finetuned model, without any inference costs and for a fraction of the time of retraining.

The Problem

AI models are not trustworthy enough for enterprise adoption

Be it LLMs or non-generative models, AI poses intolerable risks in production environments: inaccuracies, vulnerabilities, compliance issues.

The Solution

Remediate models at their core, with one click

Hirundo's unlearning eliminates the risks of AI adoption by making models forget what they shouldn't know

Select a model and what to unlearn

You can unlearn any unwanted data or behavior - anything that renders your model problematic and untrustworthy.

Get a fixed model, in less than an hour

A higher performing, unlearned model is generated in a fraction of the time it'd take to retrain it.

Optimize datasets in any modality

Aside from unlearning, our platform can increase the accuracy of any nongenerative model and dataset, detecting & remediating mislabels.

LLM Unlearning

Models can't be protected just by external guardrails. We fix AI at its core.

Unlearning allows you to remediate issues in the model itself, pushing your AI to its full potential.

Data QA for nongenerative ai

Automatically boost data and AI accuracy

Most mission-critical AI projects don't reach production. We make sure yours do. Aside from LLMs, our platform offers automated QA for vision, radar, timeseries, STT and NLP datasets and models.

Learn more
testimonials

Leading AI experts trust Hirundo

As AI regulation evolves, cost effective Machine Unlearning technology will become a must.

Avi Tel-Or

CTO, Intel Ignite

I've tried many data quality solutions. Hirundo finds data issues and mislabels at a level I’ve never seen before.

Dan Erez

AI Tech Lead, Taranis

integrations

Seamless integration with your AI stack

No workflow changes needed. Our SOC-2 certified solution runs as an API or platform, with deployment available via SaaS, VPC, or air-gapped on-premises.

Ready to forget?

Start removing unwanted data with a few clicks