Hey DEV Community! 👋
I want to share the architecture behind my live portfolio ecosystem. Instead of building a static profile site, I engineered a fully interactive experience featuring an autonomous, voice-guided AI assistant named Bunny.
👉 Live Link to Explore: roshansharma.co.in
Here is a look under the hood at how this system operates:
⚡ The Core Tech Stack
- Frontend: Next.js for high-performance rendering and smooth state machine transitions.
- Database & Vector Storage: Supabase managing a custom vector database for contextual memory.
- AI Orchestration & Chat: Autonomous LLM agent logic managing the core conversational loops.
🎙️ The Voice Guide & RAG Architecture
When a user unmutes the platform, it initializes an immersive auditory guide. The real magic happens inside the live Bunny AI Chat console on the right panel:
- Contextual Retrieval (RAG): The chat interface hooks directly into a Retrieval-Augmented Generation pipeline over my professional codebase, projects, and skills data.
- Autonomous Response Generation: Bunny can intelligently handle complex queries, offer tailored pitches, and discuss my full-stack engineering background dynamically.
- Voice Assistance Execution: The interface features a full auditory playback layer, creating a hands-free conversational workspace for reviewing engineers or recruiters.
🛠️ Reflections on the Build
The biggest engineering hurdle was balancing state synchronization between the main visual transitions, the auditory stream triggers, and the asynchronous chat retrieval states.
I’d love to know what you think of the user experience flow or the vector pipeline! Drop an unmute over at roshansharma.co.in and let's discuss down in the comments!
Top comments (3)
Overall quite cool. I had a chat with bunny and left you some comments and tips. Overall quite good, bunny froze at the end, so I couldnt confirm my last question. But overall really good. Can definitely look into your onnx voice mimicry layer, I think you overtrained it on too small a dataset. Bunny mentioned a desktop automation infrastructure? How's that coming along?
Thank you so much for this masterclass in constructive feedback, UnitBuilds! Your technical critique on the markdown text parsing bugs, the intrusive initial mic access loop, and the lack of visible data compliance frames was spot on.
Instead of a standard reply, I wanted to treat your audit with the respect it deserved. I spent the last few hours refactoring the production codebase via Antigravity, deployed the live patches, and wrote a dedicated shout-out post here on DEV detailing exactly how your insights helped me harden my architecture.
I’d be deeply grateful if you could take a quick look at the live domain (roshansharma.co.in) to see how the new on-demand mic modal and parser hold up, and jump over to the new post to drop any further suggestions or reactions. Engineers like you make building in public incredibly rewarding!
Way to go! this is really a great perspective to put AI in action. That too on a portfolio.
Keep the momentum going, love what you're building!