ML.
Projects
2025full-stack

Speech Therapy Bot

AI-powered speech rehabilitation platform delivered via WhatsApp

An AI-powered speech therapy platform for post-stroke aphasia patients, delivered entirely through WhatsApp. Uses Google Gemini for real-time speech analysis and personalized feedback, with an AdminJS back office for therapist management.

Tech stack

  • AdonisJS
  • TypeScript
  • Google Gemini
  • LangChain
  • WhatsApp API
  • PostgreSQL
  • AWS S3
  • AdminJS
  • Docker
Challenge
Post-stroke aphasia patients need consistent speech therapy practice but face barriers like limited therapist access, difficulty traveling, and unfamiliarity with new technology. The solution needed to deliver complex AI-powered speech analysis through an interface elderly patients already know.
Solution
Built the platform on WhatsApp so patients practice without learning new apps. Integrated Google Gemini via LangChain for real-time voice transcription and pronunciation scoring. Implemented adaptive difficulty levels, follow-up comprehension questions, and demonstration videos for failed attempts. Created an AdminJS back office for exercise management with CSV import and AI-generated exercise images.
Impact
Enabled patients to practice speech therapy independently between sessions using a familiar interface. The AI provides instant pronunciation feedback with 0-100 scoring, freeing therapists from repetitive assessment tasks and allowing them to focus on complex cases.