A1
Track A — Conversational & RAG-Powered Assistants

Ayurveda & Wellness Regulation Assistant

Agents that answer questions by retrieving the right information and responding conversationally, in the right language.

Background

Small Ayurveda manufacturers and startups struggle to understand which regulations — AYUSH licensing, labeling rules, IP protection for traditional formulations — apply to them. Legal consultation is expensive, and government documents are dense and mostly available only in English or formal Hindi.

Objective

Develop a multilingual AI assistant that answers questions about Ayurveda product regulation, licensing, and intellectual property — grounding every answer in a cited source document, never a guess.

Key Features

1. Conversational Q&A over regulation/licensing/IP documents 2. Source citation attached to every answer 3. Multilingual support (minimum English + 1 Indian language) 4. Confidence flagging — tells the user when it isn't sure 5. Query classification (licensing / labeling / IP / other) 6. Simple chat-style interface (web or WhatsApp-style)

Expected Solution

A retrieval-augmented chatbot backed by a document vector store A query-understanding layer that classifies intent and language A citation-grounded response generator (answers only from retrieved chunks) A translation layer for multilingual output

Suggested Technology Stack

Backend: Python / FastAPI Retrieval: LangChain or LlamaIndex + ChromaDB / FAISS LLM: Open-weight (Llama 3, Mistral) or hosted API (Claude, GPT) Translation: Google Translate API / IndicTrans2 Frontend: Streamlit / React

Expected Outcomes

Faster, cheaper access to regulatory clarity for small manufacturers Reduced dependence on paid legal consultation for basic questions A reusable RAG pattern applicable to other regulated domains

Possible Use Cases

Startup founders checking licensing requirements before product launch Manufacturers verifying labeling compliance Researchers checking IP protection status for traditional formulations

Evaluation Parameters

Answer accuracy and grounding (no hallucination) Citation quality and traceability Multilingual coverage UX clarity and response latency