MindMirror: Local-first AI reads your face & fatigue to boost digital worker productivity
Emotion detection model hits 94.5% accuracy, runs locally on your machine with Qwen LLM.
MindMirror is a local-first multimodal state-aware support system for digital workers, integrating camera-based facial expression cues, text input, optional speech interaction, structured blockage reflection, a local LLM (Ollama-hosted Qwen) for suggestion generation, and daily/weekly review reports. It forms a closed workflow: state checking, manual correction, structured articulation, suggestion generation, and state review. The fine-tuned emotion recognition model achieved 94.49% accuracy on a 6,767-image benchmark (up from 59.66%). A small formative study with six digital workers found they valued the local-first design, manual correction mechanism, and structured reflection workflow.
- Emotion recognition fine-tuned from 59.66% to 94.49% accuracy on a 6,767-image benchmark.
- Local-first design: Qwen LLM runs via Ollama, data stored in JSON/LocalStorage, optional speech API.
- Closed workflow: state check → manual correction → structured articulation → suggestion → review.
Why It Matters
An open, privacy-first alternative to cloud-based wellness tools that understands your work state without asking.