• 

  CHATBOT

• 

  code of ai chatbot

🌬️ Zephyr AI Chatbot ✨ Inspiration

The inspiration for Zephyr came from observing how most chatbots today feel either too rigid (rule-based, scripted) or too heavy (slow, resource-intensive). We wanted to build something that feels light, fast, and natural — like a breeze.

Just as the name “Zephyr” suggests, our vision was to create an AI chatbot that combines:

Effortless usability for everyday people.

Intelligent context awareness for businesses.

Customizable personalities for fun, professional, or supportive use cases.

🛠️ How We Built It

Frontend:

Built with React for a smooth chat interface.

Voice support enabled using WebRTC & Speech-to-Text APIs.

Backend:

FastAPI for handling requests and WebSocket connections.

Integrated OpenAI GPT + Hugging Face models for text generation.

Memory & Search:

Used FAISS (vector database) for semantic search and context retention.

Implemented embeddings for storing and recalling past chats.

Deployment:

Hosted backend on AWS EC2, frontend on Vercel, and used Docker for containerization.

Ensured scalability by setting up load balancing and caching mechanisms.

🚧 Challenges We Faced

Latency Issues: AI responses were sometimes delayed, so we optimized token streaming.

Context Management: Maintaining long conversations without losing coherence was tricky.

Personality Switching: Building a dynamic prompt system to let the bot switch tones (formal → funny → supportive).

Balancing Cloud & Local Models: Ensuring lightweight offline mode while keeping cloud-powered intelligence.

🏅 Accomplishments We’re Proud Of

Built a fully functional AI chatbot in a short hackathon timeframe, integrating text + voice interaction.

Successfully implemented context memory using vector embeddings, making conversations more natural.

Created dynamic personality modes (professional, casual, supportive, fun) that actually change how the chatbot responds.

Deployed the system on the cloud with scalability in mind, ensuring it can handle multiple users simultaneously.

Overcame latency challenges by enabling streaming AI responses for a smoother experience.

Designed a clean and intuitive user interface that works across devices.

Learned how to combine LLM APIs, vector databases, and real-time voice systems into one seamless product.

📚 What We Learned

Practical LLM Integration: How to combine APIs with custom datasets.

Vector Databases: Implementing embeddings for context memory.

Voice AI Challenges: Handling speech-to-text accuracy and latency.

Scalability: Designing systems that work for 1 user and 1,000+ users.

We also revisited some ML concepts, like:

Cosine Similarity ( 𝐴 , 𝐵

)

𝐴 ⋅ 𝐵 ∥ 𝐴 ∥ × ∥ 𝐵 ∥ Cosine Similarity(A,B)= ∥A∥×∥B∥ A⋅B ​

which we used to match user queries with stored embeddings.

🌍 Impact & Vision

Zephyr has the potential to:

Empower students with a personalized tutor.

Support mental health through empathetic conversations.

Assist businesses with 24/7 customer support.

Our vision is to make AI chat effortless, customizable, and universally accessible — just like a gentle breeze that adapts to everyone.

Built With

• docker
• fastapi
• google
• javascript
• postgressql
• python
• react
• vercel