AI Sponsor
Backend for a 24/7 AI sponsor chatbot for 12 step recovery programs: Claude based chat with prompt caching, Stripe subscription billing with webhooks, and a WhatsApp voice pipeline combining Whisper transcription with synthesized voice replies.
- Token cost
- 90% reduction
- Billing
- Stripe
- Tests
- 12 suites
Selected engineering work by Muhammad Mubashir, showcased by Retrieva Labs. Figures and scope reflect the existing project notes. The evidence section explains what is available to review.
A useful outcome starts here.
A recovery support SaaS needed a voice conversation channel over WhatsApp, alongside subscription billing and a chatbot backend.
From input to useful output.
- 01Receive a voice note
- 02Transcribe with Whisper
- 03Generate a contextual reply
- 04Synthesize speech
- 05Return the voice reply
The choices behind the system.
A stable prompt prefix
Static system instructions precede conversation history so prompt caching can reuse the stable part of the request.
Separate the boundaries
Session logic, webhook handling, security boundaries and audio hops need distinct tests to isolate regressions.
Observe the cost
The published note describes monitoring cache hit behavior alongside latency and token use.
Project notes, with context.
The voice pipeline article records the implementation approach and a reported input token cost reduction. Client audio, private conversations and production logs are not published here.
Read the engineering note