Multilingual WhatsApp RAG Chatbot
Budget: ₹12,500 – ₹37,500 INR
I need a complete Retrieval-Augmented Generation pipeline that lives behind a WhatsApp number and converses fluently in Hindi, English and Punjabi. End-users will either type or drop a voice memo; the agent must transcribe when needed, search the knowledge store, generate a clear answer and post it back. The primary use-case is answering general questions, so latency and language quality matter more than flashy features.
Key points I must see in your solution
• True data privacy: all messages encrypted in transit, no conversation logs stored beyond what the RAG index requires, and any third-party services configured in the most restrictive mode.
• Cost control: please favour open-source models, local or low-cost vector stores (e.g. FAISS, Chroma) and a lean hosting stack so that 10,000 monthly users can chat without the bill exploding.
• Voice support: speech-to-text for incoming memos and, if cheap enough, optional text-to-speech on replies.
• Horizontal scalability: containerised deployment or serverless design able to handle sudden peaks.
Deliverables
1. Source code, clearly documented, covering WhatsApp Business API integration, multilingual STT/TTS, vector indexing and RAG generation.
2. Setup guide that lets me recreate the full stack on my own cloud account.
3. Outline of the privacy controls you implemented.
4. A short load-test report demonstrating acceptable response times with at least 1,000 concurrent sessions (I will extrapolate to 10k).
Hand-off is complete once I can drop a voice memo in Punjabi, receive the correct answer back, and verify that no sensitive logs are left behind.
Key points I must see in your solution
• True data privacy: all messages encrypted in transit, no conversation logs stored beyond what the RAG index requires, and any third-party services configured in the most restrictive mode.
• Cost control: please favour open-source models, local or low-cost vector stores (e.g. FAISS, Chroma) and a lean hosting stack so that 10,000 monthly users can chat without the bill exploding.
• Voice support: speech-to-text for incoming memos and, if cheap enough, optional text-to-speech on replies.
• Horizontal scalability: containerised deployment or serverless design able to handle sudden peaks.
Deliverables
1. Source code, clearly documented, covering WhatsApp Business API integration, multilingual STT/TTS, vector indexing and RAG generation.
2. Setup guide that lets me recreate the full stack on my own cloud account.
3. Outline of the privacy controls you implemented.
4. A short load-test report demonstrating acceptable response times with at least 1,000 concurrent sessions (I will extrapolate to 10k).
Hand-off is complete once I can drop a voice memo in Punjabi, receive the correct answer back, and verify that no sensitive logs are left behind.