Saath
Saath is a voice-first AI medication companion for elderly Indian families, built entirely inside the Telegram chat they already use.
Core trick
A vision model reads a photographed handwritten prescription, and a managed agent checks every drug pair, then notifies family with one message.
Description
Saath is a voice-first AI medication companion for elderly Indian families, built entirely inside the Telegram chat they already use. A 68-year-old patient simply sends a photo of her doctor’s prescription, and Saath reads handwritten prescriptions in Hindi, English, or Tamil. If any field is unclear, it asks for clarification instead of guessing. Once the medication list is updated, a Claude Managed Agent runs autonomously using four custom tools: lookup_drug_info, check_pair, check_contraindication, and report_findings. It systematically checks every drug combination and every medical condition for clinically significant interactions. If no issues are found, both the patient and her daughter, often living 800 km away, receive one simple message: “No issues were found in your medicines.” If a risk is detected, such as Warfarin and Aspirin together, both phones receive a severity-graded clinical warning. The same Telegram chat also handles voice notes in Hindi, scheduled reminders with a personalized voice, smartwatch critical-alert forwarding, lab PDF parsing into plain Hindi, weekly doctor and family PDF reports, and an emergency-services regex bypass. No app installation. Just messages.
README excerpt
# Saath - V1
**Voice-first AI medication companion for elderly Indian families.**
Built for **Built with Opus 4.7** — a Claude Code hackathon, April 2026.
Team: **auraCodes**.
---
## What it does
A 68-year-old patient sends a photo of her doctor's prescription to a Telegram bot. Saath reads the handwriting in Hindi, English, or Tamil. Where a field is unreadable, Saath asks rather than guesses. Once the medication list is updated, a **Claude Managed Agent** runs in the background, autonomously checking every drug pair and every condition against documented clinical interactions. The agent decides on its own which checks to run, and when to stop.
If nothing concerning is found, both the patient AND her daughter (800 km away in another city) receive a single sentence: *"No issues were found in your medicines."* If something is found — say Warfarin and Aspirin prescribed together — both phones receive a severity-graded warning with specific clinical guidance for the next doctor visit.
The same Telegram chat handles voice notes (in Hindi or English), scheduled reminders with a personalized voice in an Indian English accent, smartwatch critical-alert forwarding to the family, lab report PDF interpretation in plain Hindi, and weekly doctor + family PDF reports.
No app. No installation. Just messages.
---
## Demo video
[Insert YouTube / Loom link here]
---
## Architecture
```mermaid
flowchart TB
subgraph Phones [Phones]
Patient[Patient phone<br/>Asha]