Inspiration
Bawk was inspired by our close friend, a behavioral technician working in Applied Behavior Analysis (ABA). He spends only 6–8 hours per week with each child, yet social skill development requires consistent reinforcement beyond those sessions. Between visits, structured practice often disappears, caregivers lack guided scenarios, and progress tracking becomes subjective. We saw a gap between clinical intention and real-world consistency — and built Bawk to extend measurable, supervised practice beyond therapy time.
What It Does
Bawk is a supervised AI-assisted social practice platform for neurodivergent children ages 7–15. Supervisors define measurable goals using structured JSON criteria, and the system converts those goals into interactive tasks like social stories, roleplay dialogues, modeling exercises, and calming routines. Children complete short assignments independently through text or voice and receive supportive, non-punitive feedback. Engagement is reinforced through XP, Pokémon rewards, and maintaining their in-app pet, turning structured repetition into consistent motivation.
How We Built It
Bawk is built with Next.js (TypeScript) for the frontend and FastAPI for secure AI orchestration. Supervisors create goals that are stored in Supabase with Row Level Security (RLS) enforcing strict data isolation. Gemini generates structured task drafts via secure server-side endpoints, and supervisors must review and publish tasks before children see them. Audio is transcribed in memory and discarded immediately to preserve privacy. Rate limiting and validation safeguards protect AI usage and maintain predictable costs.
Challenges We Ran Into
Designing for ages 7–15 required dynamic reading-level adaptation and simplified UX flows. We had to balance clinical measurability with child-friendly engagement while ensuring AI responses remained supportive and never punitive. Handling interruptions in home environments required strong resume-task logic. Additionally, controlling AI costs and preventing unsafe or unstructured outputs demanded strict server-side constraints and endpoint rate limiting.
Accomplishments We’re Proud Of
We translated behavioral therapy structures into a programmable system that keeps supervisors in control while enabling independent practice. Our goal → task → response → feedback pipeline maintains measurability and auditability. We implemented privacy-first safeguards, multi-role authentication, and structured publishing workflows that prevent unreviewed AI content from reaching children.
What We Learned
Consistency drives behavioral progress more than intensity. AI in clinical-adjacent environments must be constrained, supervised, and auditable. Structured goals with measurable criteria outperform generic practice tools. Most importantly, technology works best when it extends human supervision — not replaces it.
What’s Next for Bawk
Next, we plan to expand supervisor analytics (7/30-day trend summaries), enhance reinforcement personalization, pilot with practicing behavioral technicians, and explore compliance pathways for broader clinical deployment. Our long-term vision is for Bawk to become the reinforcement layer that sustains structured social growth between therapy sessions.
Log in or sign up for Devpost to join the conversation.