- Home
- Archives
- 2026
- Vol. 5 Issue 3
- Article
AASHA Net AI: An Offline-First, Voice-Based Health Record Assistant for ASHA Workers Using a Large Language Model
Authors
1 2 3 4 5 Department of Computer Science and Engineering, Shridevi Institute of Engineering and Technology, Tumakuru, Karnataka, India.
Article Information
- Published In
- Volume 5, Issue 3 (2026)
- Published Online
- September-December 2026
- Pages
- 926-931
- Article Type
- Review Article
Abstract
View PDFAccredited Social Health Activists (ASHAs) record health information during household visits in rural India, often without reliable internet access and usually in a regional language. Typed, connectivity-dependent mobile forms make this documentation slow and error-prone. This paper presents AASHANet AI, an offline-first, voice-based health record assistant. A Flutter application stores every record on the device first and synchronises it with a FastAPI and PostgreSQL backend through an idempotent, versioned three-step protocol that guarantees exactly-once server records under retries and detects conflicting confirmations instead of overwriting them. Voice notes in Kannada, Hindi or English are processed by the Google Gemini large language model, which returns a transcript, an English translation and schema-validated fields (patient name, age, symptoms and duration) that the health worker reviews before confirmation. The system was verified with 44 backend tests (5 against PostgreSQL), 41 mobile unit and widget tests, including 17 synchronisation fault-injection scenarios, and a live end-to-end run against the real services. Measurements showed that model selection dominates latency: on the same recording, processing took 3.0–3.4 s with one model and 96–302 s with another, which motivated configurable model fallback with time limit.
Related Articles
Share Article
*Instagram doesn't support direct link sharing from web. Copy the link and share it in your Instagram story or post.