Sentimind | Devpost

Inspiration Most of our daily communication happens online , through texts, chats, and DMs , yet we often miss what really matters: how something feels. A message can look friendly but hide toxicity, or sound blunt yet come from a place of care. That uncertainty inspired us to build Sentimind. We wanted a tool that could read between the lines , not just flag harmful language, but also highlight empathy, positivity, and trust in digital conversations.

What it does Sentimind acts as an emotional radar for online communication. It analyzes chats to detect toxicity, manipulation, or negative tone , but it also recognizes positive intent, supportiveness, and kindness. Using Daedalus Infrastructure and a custom Gemini-based LLM pipeline, Sentimind examines tone, polarity, and emotional context. It then maps patterns against Reddit’s sentiment graph via Daedalus Maps, learning how real communities respond to similar expressions. Finally, it generates an Emotional Safety Score, offering an easy-to-read snapshot of how healthy or toxic a conversation feels , without storing or exposing personal data.

Chat → Daedalus Infrastructure → LLM (analysis) → Daedalus Maps / Reddit Context → Emotional Safety Score

How we built it We built Sentimind using Flask on Daedalus Infrastructure (MCP) for fast, modular experimentation. The frontend was developed in JavaScript, with a clean visualization of the Emotional Safety Score. We also built a browser extension that lets users analyze messages instantly inside their chats. For the AI layer, Gemini powers linguistic and emotional interpretation, running securely inside a privacy-protected sandbox.

Challenges we ran into Time was our biggest challenge. We had very little time to build Sentimind from scratch, but we managed to bring it to life through tight collaboration and late-night problem solving. Integrating real-time sentiment detection into a browser extension, tuning Gemini’s responses for nuance, and ensuring privacy without losing accuracy were all tough problems to solve , but we did it together.

Accomplishments that we're proud of We’re proud that, with limited time, we built a working prototype completely from scratch as a team. It successfully identifies both toxic and positive tones in conversations and gives users real-time feedback inside the chat interface. Creating a fully functional extension connected to a Flask backend and Gemini model in such a short window was a big win for us.

What we learned We learned that emotion analysis is not just about finding toxicity , it’s also about recognizing positivity. Context matters more than keywords, and true emotional intelligence in AI comes from balance. We also learned how to coordinate under pressure, integrate multiple technologies, and still stay focused on impact and empathy.

What's next for Sentimind We plan to polish the extension and release it for public testing. Our next steps include improving accuracy for nuanced emotions and expanding integrations with popular messaging platforms. We also aim to collaborate with mental-health and safety organizations to help people recognize emotional patterns early , both good and bad. Our long-term vision is to make Sentimind the trust and empathy layer of the internet , technology that helps people communicate better, not just safer.