RIVA

• 

  Sample Image

◾Inspiration

RIVA was inspired by the idea of making club knowledge instantly accessible and interactive. In most college clubs, information is scattered across documents, seniors, and social media posts. We wanted to build a centralized, intelligent assistant for the NextGen Supercomputing Club at KIET that could answer questions naturally, not just through text, but through voice and immersive visuals. Instead of building a basic chatbot, the goal was to create an AI presence that feels dynamic, futuristic, and aligned with a supercomputing-focused community.

◾What it does

RIVA is a voice-powered AI assistant designed specifically for the NextGen Supercomputing Club. It continuously listens to user input using speech recognition, processes queries using Gemini 2.0 Flash, and responds in natural Indian English voice output. It features a dual-panel chat interface and a real-time 3D AudioSphere visualization that reacts to voice activity.

◾RIVA can:

-Answer general and technical questions -Provide detailed introductions or concise responses intelligently -Serve as a domain-specific assistant for club-related information -Deliver an interactive, immersive AI experience

◾How we built it Backend:

-Developed a REST API using Node.js and Express -Integrated Gemini 2.0 Flash for AI-generated responses -Structured prompts to maintain role consistency and knowledge boundaries -Designed backend routing to securely handle API calls and protect keys

Frontend

-Built the UI using React -Implemented a real-time 3D AudioSphere using Three.js -Used Web Speech API for browser-based speech-to-text and text-to-speech -Designed continuous listening logic with feedback loop prevention -Used React Markdown for clean response formatting -The architecture ensures that the AI model is accessed securely through the backend, keeping API keys protected.

◾Challenges we ran into

-Preventing speech recognition from picking up its own TTS output (feedback loop issue) -Managing latency between speech input, AI processing, and voice response -Designing prompts that balance general intelligence with club-specific knowledge -Handling API usage limits while maintaining responsiveness -Synchronizing real-time 3D visual effects with live audio input] -Voice-based systems required careful state management and timing control beyond a typical text-based chatbot.

◾Accomplishments that we're proud of

-Built a fully functional voice-based AI assistant from scratch -Successfully integrated AI responses with real-time 3D visualization -Designed a secure backend structure with rate-limiting considerations -Created a domain-aware assistant tailored to a real college club -Delivered a working full-stack AI product combining backend, frontend, and AI systems -RIVA demonstrates practical AI integration, not just model usage.

◾What we learned

-Building AI products requires strong system architecture, not just API integration -Voice interfaces introduce unique UX and engineering challenges -Prompt engineering significantly impacts response quality -Security and rate limiting are essential for deployable AI systems -Coordinating frontend visuals with backend AI responses demands precise state control -We learned how to transform an AI model into a usable, domain-specific product.

◾What’s next for RIVA

-Implement authentication for club members -Add persistent memory and conversation history -Expand the knowledge base with structured club data -plan to give it a face , 3d avatar Deploy a scalable version with controlled API management iot intergration

Built With

• express.js
• gemini
• node.js
• react-native
• three.js
• webspeech