Wellnest

An AI health companion you can talk or text on web, mobile, and Telegram that remembers your conversations, answers health questions, analyzes symptoms, and recommends items to help.

Comment

Inspiration

Managing recurring health concerns takes effort long before a medical appointment. Questions arise between visits, symptoms are easy to forget, and searching online can leave people with more information but less clarity.

We envisioned a health companion that fits into everyday life: somewhere to ask questions, build on previous conversations, and explore next steps through an app or a simple Telegram message.

What it does

Wellnest is an AI health assistant on web, mobile, and Telegram that you can text or talk to.

  • Conversational support for health questions, by text or by voice.
  • Voice mode, on by default, that transcribes what you say and speaks its replies. On Telegram, it answers voice notes with a voice note of its own.
  • Memory that retrieves relevant details from a user’s previous chats.
  • Web search and biomedical research tools for questions that need more information.
  • An urgency-screening step that flags potentially serious concerns and encourages timely professional care.
  • A symptom analysis that runs only with the user’s consent. It ranks the most likely conditions and lists red flags and care guidance.
  • Product recommendations for supplies that can help with the user’s condition.

Wellnest supports health understanding and access to information. It is not a diagnostic service or a replacement for professional medical care.

How we designed it

Our architecture uses React Native with Expo for a shared web, iOS, and Android codebase, with Convex handling backend functions, message storage, and incoming Telegram webhooks.

When a message arrives, Jina generates an embedding, and Convex vector search retrieves relevant past conversations for that user.

Before the core model answers, Jev screens each message. It checks for urgency, decides whether supplies would help, and confirms whether the user agreed to a symptom analysis. We use Jev for this step because its low latency matters for time-sensitive issues.

Snowflake Cortex is the reasoning layer that writes the responses, and a CREAO agent runs the symptom analysis in the background.

CREAO agent (Symptom & Biosensor Health Analyzer):

Cartesia powers voice: speech-to-text on the way in and streamed speech on the way out.

Telegram connects to the same backend pipeline, so users can reach Wellnest through a familiar messaging app.

Challenges

The central design challenge is making multiple interfaces feel like one continuous conversation. App and Telegram messages need to share useful context while keeping each user’s information separate.

Another challenge is balancing helpful health information with uncertainty. A conversational answer must not imply a confirmed diagnosis, and urgency screening must not be presented as a guarantee that every emergency will be detected.

Another challenge that we faced was that a lot of public retailers banned AI shopping so we remedied this by instead of having the AI shop for you it just recommends items that the user could purchase.

We also had to keep the scope achievable. We focused the MVP on conversation by text and voice, memory, and research, leaving external health-data integrations for future development.

What we learned

Through the making of Wellnest we learned how to make Telegram agents.

It also highlighted the value of a shared backend pipeline. The same processing flow can support messages from the web app, mobile app, and Telegram.

What we accomplished

We built an assistant that works in the app and on Telegram, by text or by voice. When you describe your symptoms, it can give you a rundown of possible conditions and steps to help you feel better. If certain products could help, it recommends those too, and for serious issues it urges you to see a physician. We tested the agents against real symptoms and edge cases.

What's next for Wellnest

Our next priorities are evaluating response quality, testing urgency handling, and giving users more control over their history, building on the new Clear chat option.

Longer term, we hope to add opt-in integrations with more health apps and to expand shopping help for everyday health supplies.

Built With

Share this project:

Updates

Submission history