README

title: GOFA AI Overview sidebar_label: Overview description: Overview of GOFA AI intelligent services and RAG Chatbot API grouping.

GOFA AI provides intelligent health assistant capabilities centered on Retrieval‑Augmented Generation (RAG). All responses are required to be grounded in retrieved content, avoiding hallucination.

The first publicly available endpoints are the RAG Chatbot API (/api/rag-chatbot) and its accompanying text-to-speech endpoint (/api/rag-chatbot/tts). This section will:

Quick Navigation

• RAG Chatbot API – Interactive retrieval-augmented health consultation
• Upcoming features (planned): Personalized risk summaries, adaptive rehabilitation suggestions, explainability metrics, etc.

Core Design Principles

Principle	Description
Grounded Responses	Answers must cite retrieved document content, no speculation
Language Adaptation	Chinese input responds in "Cantonese colloquial"; English input responds in concise professional English
Safety Compliance	Each response includes a medical disclaimer, not a substitute for professional advice
Scope Restriction	Non-health-related questions are politely redirected and constrained
Traceability	Responses end with source document titles listed as references
Streaming First	Default to SSE streaming output for improved perceived response speed

Endpoint Overview

Category	Endpoint	Description
Chat Completion	POST /api/rag-chatbot	RAG streaming/single-shot responses (with source sentinel)
(Accompanying) TTS	POST /api/rag-chatbot/tts	Synthesizes speech from response text (explained separately)

Recommended Integration Flow (High-Level)

1. The client first obtains a Firebase ID Token (corresponding to your user).
2. Send POST /api/rag-chatbot, preferably without ?format=json to experience streaming.
3. The frontend concatenates text-delta incrementally, waits for the final source sentinel => separates answer and sources.
4. If speech is needed, call the TTS endpoint after obtaining the final clean text (removing the sentinel segment).
5. Store the answer, source list, model, and timestamp in conversation records (to avoid re-parsing raw text).

Versions and Compatibility

Component	Status	Notes
Model	gemini-2.5-flash	May adjust with GCP upgrades, check release notes
Source Sentinel	Stable	Pattern: \n[__SOURCE_FILES__] + JSON array
JSON Mode	Stable	Suitable for batch, export, or atomic result scenarios

Related Documents

• RAG Chatbot API Detailed Specifications

---

For new features or adjustments, please include use cases and expected behavior when submitting an issue to facilitate evaluation.