AIxBlock

AIxBlock is an enterprise training-data company for speech and large language models, backed by the European Union's innovation funding programme. It supplies speech collection and transcription, conversation and dialogue annotation, intent and entity labeling, RLHF-style preference data used to align models and reduce hallucinations, and an off-the-shelf library of hundreds of thousands of hours of real-world call-centre audio covering US, Indian, and Philippine accents alongside Indian languages. It works in roughly 100 languages and reports more than 200,000 hours delivered. Its pitch to regulated banking, healthcare, and government buyers is data sovereignty: clients connect their own storage so AIxBlock never holds a copy and cannot resell the data. The company's history needs unpacking, because the widely quoted "7 years" blends two entities. Its founders started a first AI-training-data company in 2019, delivered large-scale projects for Fortune 100 firms and unicorns including Oracle, AWS, Apple, Uber, Samsung, Uniphore, and Chainalysis, and exited that business in 2023. AIxBlock itself launched in 2023 with EU backing, spent two years building AI infrastructure, and repositioned in 2025 around enterprise speech and LLM data. So that Fortune 100 client list traces mainly to the predecessor company rather than to AIxBlock's own trading record. CEO and co-founder Ha Dao is publicly identified as a former co-founder of Wow AI, matching that 2019 company and its exit. One further piece of history is worth knowing because it still dominates search results: during its 2023 to 2025 infrastructure phase AIxBlock was positioned as a decentralized Web3 AI platform on Solana, where contributors accrued points convertible into tokens at a future Token Generation Event. The current contributor offer is conventional, with payment by bank transfer, Payoneer, or stablecoins, so confirm present terms directly rather than relying on those older write-ups.

AIxBlock recruits contributors worldwide and says it draws from more than 100 countries, which makes it one of the more geographically open platforms in this category. The work falls into four types: recording and transcribing audio in your native language, annotating text and dialogue by labeling conversations and tagging intents and entities, collecting audio including voice recordings and environmental or machine sounds to project specification, and AI evaluation and RLHF work comparing and ranking model outputs. It explicitly wants native speakers of any language, transcriptionists and linguists, domain experts in medicine, law, finance and tech, detail-oriented annotators, and voice actors or audio recording professionals, and it provides hands-on training and guidelines, so prior experience is not strictly required. Getting started is a short application covering your skills and languages, after which you browse available projects, apply, take a test, and onboard. Live listings observed include a worldwide code-switching recording project, a 9,900-video face-motion project with separate freelancer and vendor-partner tracks, robotic video collection, and a US-only HCC risk-adjustment medical coder role, so eligibility varies sharply by project. Two things to go in knowing. Per-project rates are not published publicly, unlike some peers, so you have to confirm pay per project before committing time. And verification is unusually heavy: KYC, biometric enrollment, device fingerprinting and credential validation up front, then random biometric re-authentication and liveness checks during sessions, plus behavioural anomaly detection. That protects honest contributors from proxy fraud, and it is also genuine biometric data collection, so weigh it accordingly.