Next.jsTypescriptGroqllama.cppSpeech-to-text
The Problem
Most digital applications compete for users' attention and encourage more screen time. People often overlook everyday details in their surroundings without actively practising observation.
The Solution
Observation Coach takes a different approach. Instead of using AI to identify everything for the user, it encourages people to observe their surroundings independently and uses AI to guide their reflection through meaningful follow-up questions and adaptive exercises.
Key Features
- Guided Observation Sessions — Choose 5, 10, 20, or 30-minute observation exercises.
- AI-Powered Coaching — Receive contextual follow-up questions based on your observations.
- Adaptive Exercises — Get subsequent exercises based on your previous reflections.
- Observation Progression — Practise attention, description, comparison, change detection, and inference.
- Local AI Inference — Run open-weight Qwen models locally using llama.cpp.
- Hosted AI Support — Switch to hosted inference through OpenAI-compatible APIs.
- Guided Demo Mode — Explore the application without configuring an AI API key.
Technical Highlights
- Designed a modular AI provider architecture supporting both local and hosted inference.
- Integrated Qwen 3B Instruct (GGUF) through llama.cpp for local inference.
- Configured hosted inference using Qwen 27B Instruct through Groq.
- Implemented structured AI responses with validation before application-level processing.
- Used MongoDB for user accounts, observation sessions, saved observations, and progress tracking.
- Built an adaptive coaching workflow that encourages independent observation rather than automated object recognition.
Tech Stack
Next.js · TypeScript · Tailwind CSS · MongoDB · Qwen · Groq · llama.cpp · Hugging Face
Project Demo
- Product Walkthrough: https://youtu.be/KArn5bRtpeI
- Code Walkthrough: https://youtu.be/vQHZ7amHMTo
Built For
Hacktoberfest 2026 — Open-Source AI Challenge, Week 1: Touch Grass.