Reflection

Inspiration

Most people have goals, but staying consistent is hard. Coaches help, but they’re expensive and not always available when you need them most. We wanted to build a system that integrates coaching into daily life without adding friction or requiring another app. Reflection is inspired by the idea that a mirror is the first thing you see every day — so why not make it a place where you can get personalized guidance, motivation, and accountability?

What it does

Reflection turns a smart mirror into a personal coach. When a user taps an NFC totem, the mirror recognizes them, loads their goals and context, and starts a voice-based coaching session. The coach listens, tracks progress, and provides tailored advice and next steps. Over time, it learns the user’s preferences and adapts its coaching style to fit their personality and goals.

How we built it

Reflection is built as a hybrid system combining a modern UI with an AI-powered backend and NFC-based identity verification.

Frontend (Mirror UI)

Built with React + Electron + TypeScript
Displays goals, progress, and coaching prompts
Handles real-time interaction and UI updates

Backend (Logic & Intelligence)

Implemented in Python
Manages NFC authentication, goal tracking, and AI integration
Communicates with the frontend via WebSockets for real-time responses

Database (Memory)

MongoDB Atlas stores:
- User profiles
- NFC mappings
- Goals
- Long-term context
Each NFC totem maps to a unique user profile, enabling instant personalization

AI & Voice

Whisper for speech-to-text
LLMs + RAG for intelligent coaching and long-term memory
ElevenLabs for natural text-to-speech

Hardware

Webcam and microphone capture input
Arduino with a PN532 NFC module handles identity scanning
Mirror display provides visual feedback and coaching prompts

Challenges we ran into

Consistent coach behavior: Making the AI always act like a coach required careful prompt engineering and context control.
Identity reliability: Ensuring NFC scanning consistently loaded the correct user profile every time.
Real-time voice interaction: Building a seamless loop for speech-to-text, AI processing, and text-to-speech with minimal latency.
Context management: Storing and retrieving long-term user context without sacrificing performance required a robust RAG design and database structure.

Accomplishments we’re proud of

Implemented a reliable NFC identity system that instantly loads user profiles and goals.
Built a real-time voice coaching loop using Whisper, LLMs, and ElevenLabs.
Designed a RAG memory system that enables long-term personalization and consistent coaching.
Developed a polished mirror UI that provides a smooth user experience and clear goal visualization.

What we learned

Reflection taught us that AI works best when paired with structured memory. Without context, AI coaching can feel generic or inconsistent. We learned that a coach needs a stable identity and long-term memory to be truly helpful. We also realized how important user experience is — the system must feel fast, reliable, and non-intrusive to become part of daily routines.

What’s next for Reflection

Next, we plan to:

Improve personalization through fine-tuning
Add goal analytics and progress visualization
Expand into corporate mode for team coaching and shared goals
Explore offline functionality by running models locally on the mirror hardware to improve speed and privacy