AI Mental Wellness Chat Platform -- 2
Budget: ₹750 – ₹1,250 INR
I’m building a student-centric mental-wellness platform and the first milestone is the AI-powered “emotional first aid” feature. The goal is to let any college student open a text chat, type how they feel, and receive real-time, empathic guidance generated by an NLP model. The system should
• read each incoming message,
• run sentiment and risk scoring on the fly,
• suggest personalized coping strategies or grounding exercises, and
• trigger an escalation flow if the model detects crisis language (routing to a human peer supporter or the counselor-booking screen that already exists in the wider product).
Key points to keep in mind
– Analysis must be fast enough to feel live; latency under two seconds per turn is the target.
– Text only for now; voice or video analysis can remain future roadmap items.
– All responses need a transparent “Why we suggested this” note so the user and our compliance team understand the AI’s reasoning.
– Please use mainstream, well-supported NLP tooling (e.g., Python + spaCy or a transformer service such as OpenAI / Claude), wrapped as REST or GraphQL endpoints so our React front end can consume them.
Deliverables for this milestone
1. Model and inference code with clear instructions for fine-tuning or prompt management.
2. Real-time chat API (WebSocket or Server-Sent Events preferred) with sentiment & risk scores returned alongside replies.
3. Simple admin dashboard page that lists flagged high-risk conversations.
4. Setup docs and Dockerfile so our DevOps team can deploy to AWS or GCP without surprises.
I’ll provide Figma flows, brand voice guidelines, and the existing user-auth API once we kick off. Let me know which stack you propose, any prior work on conversational mental-health tools, and an estimated timeline to hit the four deliverables above.
• read each incoming message,
• run sentiment and risk scoring on the fly,
• suggest personalized coping strategies or grounding exercises, and
• trigger an escalation flow if the model detects crisis language (routing to a human peer supporter or the counselor-booking screen that already exists in the wider product).
Key points to keep in mind
– Analysis must be fast enough to feel live; latency under two seconds per turn is the target.
– Text only for now; voice or video analysis can remain future roadmap items.
– All responses need a transparent “Why we suggested this” note so the user and our compliance team understand the AI’s reasoning.
– Please use mainstream, well-supported NLP tooling (e.g., Python + spaCy or a transformer service such as OpenAI / Claude), wrapped as REST or GraphQL endpoints so our React front end can consume them.
Deliverables for this milestone
1. Model and inference code with clear instructions for fine-tuning or prompt management.
2. Real-time chat API (WebSocket or Server-Sent Events preferred) with sentiment & risk scores returned alongside replies.
3. Simple admin dashboard page that lists flagged high-risk conversations.
4. Setup docs and Dockerfile so our DevOps team can deploy to AWS or GCP without surprises.
I’ll provide Figma flows, brand voice guidelines, and the existing user-auth API once we kick off. Let me know which stack you propose, any prior work on conversational mental-health tools, and an estimated timeline to hit the four deliverables above.