GlowAI

Inspiration

We developed Glow because we care about our health but it is difficult to actually know what is good and what is bad for our bodies. We focused on diet, exercise + sleep, and skin as these were the main categories that we found difficult to optimize ourselves.

Amidst busy schedules, endless product options, and too many harmful ingredients, Glow seamlessly tracks our lifestyle and provides clear guidance on meaningful improvements.

What it does

The Glow diet section tracks your diet and rates the foods you eat by analyzing their ingredients. Adding meals is made simple by providing barcode scanning for millions of products. The sleep and exercise page tracks the amount of exercise and sleep health. Finally, skin care routine and products can be tracked and rated via the skin health section.

This all sounds complicated until you realize: It can all be done by speaking to the agent for a minute a day. The specialized agents intelligently pick up on key points during the conversation and track details automatically.

The app summarizes all your results into a single value: a Glow score, which tracks your ongoing progress on a scale from 1-100 and encourages you to continue making incremental progress.

How we built it

Frontend: TanStack Start + React + TypeScript + Vite for the web app, logging UI, barcode/search flows, voice input, and Glow Score dashboard. Backend: Supabase Edge Functions run the agent workflows and API orchestration. Brain: Fetch.ai ASI:One powers the LLM reasoning for skincare, diet, sleep/recovery, and orchestration. Agents: Skincare Agent, Diet Agent, Wellness Agent, and Glow Orchestrator analyze category-specific inputs and merge them into one Glow Score. Data: Supabase Postgres stores profiles, onboarding answers, daily logs, skincare history, meal/sleep/exercise logs, ingredient analyses, and score history. Product Intelligence: Open Food Facts and Open Beauty Facts populate food/skincare product data; EWG, CosIng, and INCI references provide ingredient safety context. Voice: Web Speech API handles speech input; browser TTS handles spoken responses. Scoring: Agent outputs are normalized into a live Glow Score with category breakdowns and historical tracking.

Challenges we ran into

Agentic platforms, time based constraints, etc. Connecting various APIs and deploying the final web app to seamlessly run was also quite difficult.

What we learned

We learned a lot about agentic flow and how to quickly develop ideas using Vercel. It was a learning curve but working together on the same project in a short time frame taught us how to brainstorm under pressure and break apart large features into buildable elements.