An-tho
An-tho is an independent applied AI lab focused on building trustworthy AI applications through rigorous engineering. Its first project, Vela, was a multilingual medical research tool that provided cited, evidence-based answers from PubMed and FDA drug data across 16 languages, with features for drug interaction checks and lab report explanations. Vela was archived in October 2026, though its research functionality remains available for demonstration.
- Artificial Intelligence
- Healthcare Technology
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
Healthcare professionals who work across languages face significant barriers when searching medical literature, as most AI search tools are optimized for English and fail to provide verifiable, source-grounded answers in other languages. This limits access to peer-reviewed evidence and official drug labels, potentially impacting clinical decision-making in multilingual and non-English settings.
Solution
An-tho is an independent applied AI lab that builds trustworthy AI applications through rigorous engineering. Its first project, Vela, was a multilingual medical search tool designed specifically for healthcare professionals working beyond English. Vela allowed users to ask questions in any language and receive answers in their own language, with every response citing peer-reviewed literature and official drug labels. The tool clearly flagged content that the model generated on its own, distinguishing it from sourced information, and noted where local guidance may differ from cited sources. Vela required no identity or license verification and was anonymous by default, with user data never shared, sold, or used to train AI. The tool was archived in October 2026, though its research functionality remains available for demonstration.
Target Audience
Primary users are healthcare professionals who work in multilingual or non-English clinical environments and need reliable, source-verifiable medical information search.
Features
- Multilingual search across PubMed's 36M+ articles and official FDA drug data, with answers returned in the user's language across 16 supported languages
- Source-first design where every answer includes citations, and model-generated content is explicitly flagged
- Local guidance discrepancy alerts that indicate where regional medical guidance may differ from cited sources
- Drug interaction verification powered by FDA OpenFDA with structured severity ratings (Critical / Major / Moderate / Minor)
- Lab report explanation that identifies entities via LOINC, RxNorm, and MedlinePlus, generating plain-language explanations with source badges
- Security layers including PHI detection (TW/JP/US patterns), prompt injection guards, intent classification, and per-IP rate limiting
- Anonymous by default with no identity or license verification required, and no data sharing, selling, or use for AI training
- Research demo still accessible after the product's October 2026 archival