Inspiration:
Millions silently battle anxiety, guilt, and burnout but can’t access therapy—due to stigma, cost, or fear of judgment. SOUL-INK was born to give them an anonymous, empathetic AI listener that actually understands what they feel and helps them translate chaos into clarity.
What it does:
SOUL-INK lets users speak or type their thoughts. It detects emotional tones with NLP (GoEmotions), asks Socratic questions to guide reflection, and offers real, short coping mechanisms—like grounding or journaling micro-tasks. Everything is stored only in memory (session-based), so users can export or delete their reflections anytime. A minimal dashboard visualizes emotional patterns and recurring themes.
How we built it:
Frontend: Streamlit (custom neon-purple 6dainn aesthetic) NLP: Transformers (DistilBERT GoEmotions), spaCy for trigger extraction Audio: SpeechRecognition + pydub/ffmpeg for voice uploads Visualization: Plotly interactive dashboards Deployment: Google Colab + ngrok tunnel or localhost Streamlit run
Challenges we ran into:
Balancing depth with privacy (no database, no logging) Handling noisy speech inputs & transcription accuracy Keeping inference lightweight enough for free GPU/CPU sessions Designing Socratic responses that feel human yet safe
Accomplishments that we're proud of:
A working anonymous AI reflection companion in under 500 lines of code Smooth voice-to-text pipeline integrated with emotion analysis Fully local session—no data stored, ever Design that merges art with empathy and AI
What we learned:
Emotion understanding with small transformer models can feel surprisingly authentic “Therapeutic” design isn’t about medical advice—it’s about listening structure Building human-centric AI requires tone, pacing, and boundaries—not just accuracy
What's next for SOUL-INK:
Browser-native speech capture (no upload needed) Multi-language emotion support (Arabic, French, Spanish) Daily reflection streaks + journaling rewards Offline desktop build using Electron + local LLM inference
Log in or sign up for Devpost to join the conversation.