R2D-Talk

Inspiration

We recognize that small businesses lose revenue simply because they aren't able to answer all incoming calls. Yet, for many, hiring a receptionist or outsourcing an answering service is too expensive. We believe there is a better way.

What it does

R2D-Talk automates your small business's reception, allowing you to answer incoming calls you would have missed. Trained on your business data — including company PDFs and Word Docs (e.g. Restaurant menus), Databases and Google Apps (e.g. Restaurant reservations system), and even business website content — we provide a conversational phone agent that does all the work of a receptionist.

How we built it

1. Data → Vector DB: Upload business data and we’ll embed it.
2. LangChain: We use retrieval-augmented generation with LLMs to augment responses with live business database data and unstructured documents/PDFs
   
3. Text ↔ Speech: With Twilio, we convert text to speech and vice-versa to create a conversational phone agent that can deploy the LangChain LLM.

Our data is collected with BeautifulSoup and PostgreSQL with PyPDF and PythonDocx. Our backend is based in a microservice architecture with Flask. Our Vector store utilizes LangChain and FAISS. Our text-to-speech from Vosk integrates into Twilio and connects to Google Calendar API.

Accomplishments that we're proud of

• We have a fully working live demo and commercializable use case with integrations
• We were able to create a high-impact product that could be relevant for multiple industries
• Lasya is a first-time hacker and learned a lot

Built With

• amazon-web-services
• beautiful-soup
• elevenlabs
• gpt
• langchain
• python
• twilio
• whisper