Architecting a Yoga Studio Chatbot: Multi-Model Routing and Low-Cost Memory
omi0009 · reddit · 2026-08-03
A developer shared their architecture and model selection experience for building a production-grade chatbot for a yoga studio in Vietnam. The app handles appointment booking, empathetic coaching, wellness advice, and FAQs. It serves 200 active users with 6k-8k messages/month, aiming for a budget under $5/month while requiring cross-session memory.
Core Architecture & Selection:
- Intent Routing: User messages hit a FastAPI webhook, fetch history from MongoDB, and route by intent.
- Multi-Model Setup: Uses Llama 3.3 70B for deep therapy, GPT-OSS 20B for strict JSON booking/tool calling, and Llama 8B for general FAQs.
- Memory Mechanism: Short-term memory uses the last 20 messages; long-term memory relies on profile summaries, evaluating the need for a vector DB (Pinecone/Qdrant) for semantic search.
- Resilience: Implements exponential backoff for rate limits, a model fallback chain (70B -> 20B -> 8B), circuit breakers, and DLQ for failed calendar syncs. Prompt injection guards and PII masking are also included.
More from coding & agent
- Migrating from GPT-4o to GPT-5.1: Handling RAG Agent Response Style Regressions — IncreaseLocal2574 · 2026-08-03
- Using MCP to Seamlessly Invoke GlobalGPT in Codex for Writing and Video Generation — thetripathi58 · 2026-08-03
- Open-Source Skill Watches Videos and Generates Structured Notes — Roger_M_Taylor · 2026-08-03
- Java Dev Builds AI Agent in 30 Minutes Without Python Using LangChain4j — deepakatl1981 · 2026-08-03
- OpenRoboto Robotics Challenge: Fine-tune π0.5 to Win Token Rewards — const_reborn · 2026-08-03
- GitHub Project OpenCode: Orchestrating 7 Specialized Agents for Parallel Coding Tasks — tom_doerr · 2026-08-03