Hirundo

The company automatically debugs training data to improve accuracy in vision, radar, time-series, tabular, or STT models.

Showcasing Hirundo’s bias unlearning capabilities on DeepSeek-R1-Distill-Llama-8B, reducing its biases up to 76% compared to original performance.

VinDr-CXR is one of the largest and most trusted public datasets of chest x-ray scans (CXR). It features over 18,000 CXRs, annotated by 17 radiologists with at least 8 years of professional experience

Hirundo's Machine Unlearning Platform offers a comprehensive solution for selectively removing unwanted data from various AI models.

Hirundo has achieved an 86% reduction in prompt injection vulnerabilities and a 78.6% drop in biased behaviors in Cohere’s Aya-Expanse-32B

The company removes unwanted data from Large Language Models without requiring retraining or guardrails, including PII, noncompliant, copyrighted, or confidential information.

The company helps users remove unwanted data from AI models and training datasets for optimization, debugging, and compliance.

Hirundo, an Israel-based company, develops high-resolution model editing tools to mitigate hallucinations, bias, and security vulnerabilities in large language models (LLMs). CEO Ben Luria explains the company’s redaction-based platform, which targets and removes unwanted behaviors in AI systems to improve accuracy and reliability. The approach focuses on refining LLMs for safer, more trustworthy applications.

Aug 2025Center for Data InnovationNews article

Hirundo wins first place in the RAISE Summit Startup Competition, defeating over 1,100 AI startups. The company’s technology allows enterprises to modify AI models to exclude problematic data and behaviors. Judges from OpenAI, Nvidia, Google, and Sequoia evaluated submissions in the prestigious contest.

Jul 2025PR NewswireNews article

Hirundo, an Israeli cybersecurity startup, secures $8 million in seed funding to address AI hallucinations and bias. The company aims to help organizations deploy safer AI models by mitigating these risks. The funding supports development of tools targeting unreliable AI outputs and discriminatory patterns.

Jun 2025SecurityWeekNews article

Hirundo secures $8 million in seed funding to develop machine unlearning solutions for AI hallucinations, bias, and data vulnerabilities. The round includes investors like Maverick Ventures Israel, SuperSeed, and Alpha Intelligence Capital. The startup focuses on addressing core AI reliability challenges through specialized techniques.

Jun 2025Unite.AINews article