Bobidi

Bobidi operates as a testing platform for artificial intelligence models, enabling companies to validate their AI systems before deployment. The firm was established in June 2021 by co-founders Jeong-Suh Choi and Soohyun Bae, who previously held positions at Facebook and Niantic, respectively. Choi's background includes experience as a Product Manager at both eBay and Facebook, as well as involvement in multiple startups. The founding team's collective experience spans major technology companies like Meta (formerly Facebook), eBay, Google, and Niantic. Bobidi secured $5.8 million in a seed funding round on July 14, 2022, with participation from investors including Y Combinator, Meta, and Hyundai Motors Group.

The company's core business revolves around addressing the challenge of identifying and managing uncommon situations or "edge cases" in AI models. Bobidi provides a platform that connects a client's AI model to a global community of human testers via a secure API. These testers work to uncover biases and flaws, effectively acting as a bug bounty community for AI. This process aims to shorten the model validation timeline from months to weeks or even days. The platform is designed to be more efficient than traditional methods by focusing on collecting novel data that the model specifically needs to improve. For generative AI clients, the service helps refine training datasets for quality and reduce inference costs by using programmatic filters to scan entire datasets and remove harmful data points related to toxicity, bias, and misinformation.

Bobidi's business model appears to be usage-based, where clients pay based on the number of legitimate testing attempts made by the community. One reported pricing structure is $0.99 for ten "legit" attempts. The platform also creates an earning opportunity for its global community of testers, who can reportedly make between $10 and $20 per hour, a rate that is often significantly higher than local wages in many countries. The company serves clients in sectors such as augmented reality and computer vision. Initially focused on the AI testing platform, some sources also mention a service called Upswell, which creates quiz-based loyalty campaigns for local restaurants from their promotional videos.

Keywords: AI model testing, bug bounty for AI, AI validation, data quality, edge case detection, AI bias testing, generative AI data refinement, model validation platform, human-in-the-loop AI, crowdsourced AI testing, dataset quality, AI safety, machine learning validation, Jeong-Suh Choi, Soohyun Bae, Y Combinator, AI data services, reduce AI bias, AI performance, data-centric AI