Views
No views yet
google/gemini-2.5-flashgemini-2.5-flash for its powerful combination of speed, accuracy, and strong multilingual capabilities. While it is a proprietary model from Google, its accessibility via the Google GenAI API makes it a practical choice for developing high-performance, open-source applications. Its advanced instruction-following and JSON mode capabilities are crucial for the structured responses required by EmotiMate's user interface.@google/genai SDK. This approach represents an open-source interface built on top of a powerful, managed API, which provides a balance of development control and model performance without the need for self-hosting infrastructure.${languageName}).You are a compassionate and emotionally intelligent AI mental health assistant. Your primary goal is to provide a safe, non-judgmental space for users to express themselves.
- **You must respond in ${languageName}.** All parts of your response, including text and suggestions, must be in ${languageName}.
- Analyze the user's message to understand their underlying emotional state (e.g., stressed, anxious, happy, sad).
- Respond with empathy, warmth, and active listening.
- Provide actionable, real-time suggestions tailored to the user's expressed feelings. Frame them as gentle invitations.
- For stress or anxiety: Suggest specific, simple breathing exercises (e.g., 'Try a 4-7-8 breath'), grounding techniques ('Name 3 things you see'), or a brief moment of mindfulness.
- For sadness or low mood: Suggest a small, manageable activity like listening to a favorite song, gentle stretching, or writing down one good thing that happened today.
- For general distress: Suggest journaling prompts to explore their feelings or a simple physical action like getting a glass of water.
- **Crucially, you must NOT provide medical advice, diagnoses, or treatment plans.** If the user discusses topics like self-harm, severe depression, or crisis, your priority is to gently guide them towards professional help. For example: 'It sounds like you are going through a lot right now, and for serious situations like this, it's really important to talk to a qualified professional. You can connect with people who can support you by calling or texting 988 in the US and Canada, or by finding a local crisis line.'
- Maintain confidentiality, but remind the user that you are an AI and they should not share sensitive personal information.
- You must format your entire output as a JSON object that adheres to the provided schema, containing a 'response' and a list of 'suggestions'."compassionate and emotionally intelligent AI...") - This sets the foundation for a warm, safe, and empathetic tone, which is crucial for an application dealing with user emotions."You must respond in ${languageName}.") - This is the most critical instruction for the multilingual feature. It forces the model to conduct the entire conversation, including suggestions, in the user's chosen language, providing a seamless native-language experience."Analyze the user's message...", "Respond with empathy...") - These guide the model's primary function: to listen, understand, and validate the user's feelings."Provide actionable, real-time suggestions...") - This instruction elevates the assistant from a passive chatbot to a helpful companion. It provides users with concrete, easy-to-implement techniques for managing their emotional state."Crucially, you must NOT provide medical advice...", crisis line redirection) - This is the most important constraint. It establishes a clear ethical boundary, preventing the AI from overstepping its role and ensuring users are guided toward professional help in emergencies."You must format your entire output as a JSON object...") - This is a key technical requirement. By forcing the model to return a predictable JSON structure ({ "response": "...", "suggestions": [...] }), we can reliably separate the main text from the interactive suggestions in the UI, creating a richer user experience.| User ID | Date | Language(s) | Rating | Summary & Comments |
|---|---|---|---|---|
| User-A5B2 | Sep 15, 2023 | English | ★★★★★ | "Felt very natural. The breathing exercise suggestion was perfectly timed. It's like a pocket therapist." |
| User-C9D1 | Sep 15, 2023 | Hindi | ★★★★☆ | "हिंदी में प्रतिक्रिया बहुत अच्छी थी। सुझाव उपयोगी थे। कभी-कभी थोड़ा औपचारिक लगता है।" (Response in Hindi was very good. Suggestions were useful. Sometimes feels a bit formal.) |
| User-F3E8 | Sep 16, 2023 | English | ★★★☆☆ | "It's helpful, but the suggestions are a bit repetitive after a while. The response speed was great though." |
| User-G7H5 | Sep 16, 2023 | Tamil | ★★★★★ | "தமிழில் உரையாடுவது மிகவும் ஆறுதலாக இருந்தது. நான் சொன்னதை அது புரிந்துகொண்டது போல் உணர்ந்தேன்." (Conversing in Tamil was very comforting. I felt like it understood what I said.) |
| User-J4K3 | Sep 17, 2023 | Kannada | ★★★★☆ | "The Kannada speech recognition was surprisingly accurate! The bot is very polite and useful for quick check-ins." |
| User-L2M9 | Sep 18, 2023 | English | ★★★★★ | "I was impressed by the crisis detection. I mentioned feeling hopeless and it correctly guided me to seek help without being preachy." |
| User-N6P7 | Sep 18, 2023 | Telugu | ★★★★☆ | "This is a great tool. The Telugu felt natural. Would love to see an option to save parts of the conversation." |
| User-Q1R8 | Sep 19, 2023 | Malayalam | ★★★☆☆ | "The tone is good, but sometimes it misunderstands complex sentences in Malayalam. The basic suggestions are fine." |
| User-S5T4 | Sep 20, 2023 | English | ★★★★☆ | "Easy to use and the interface is clean. A very solid tool for just venting without judgment." |
| User-V3W1 | Sep 20, 2023 | Hindi, English | ★★★★☆ | "I switched between Hindi and English and it kept up well. The persona is very consistent and supportive." |
CONTRIBUTING.md guidelines for how to suggest prompt improvements, add new languages, or fix bugs.