Backend-focused developer | B.Tech CSE student at PES University
I like building systems that actually run end-to-end — from database schema to API to the thing a user taps on their screen. Most of my recent work has been around agentic AI, real-time systems, and full-stack product builds.
A local-first, voice-controlled AI assistant built as a capstone/research project. VIMA listens, thinks, and talks back — with tool-calling for real tasks like contacts, email, and messaging.
- Architecture: Electron desktop app with a Node.js/Express orchestrator, a Python FastAPI microservice for Whisper (speech-to-text) and edge-tts (text-to-speech), and a React/TypeScript frontend
- Real-time: WebSocket-first communication — binary audio streaming in, generated audio streaming back out
- LLM routing: Hybrid local/cloud model routing (Ollama for local inference, Gemini as a cloud fallback) with automatic online/offline detection
- Agent design: ReAct-style orchestration, fuzzy contact resolution (Google People API + LLM disambiguation), and confirmation-gating before any destructive tool call runs
- Data: SQLite + Prisma for conversations, messages, and per-user settings
GitHub: github.com/PuneethV333/vima-bot Release: v1.0.0
A Node.js-based scaffolding tool that creates scalable frontend and backend projects with multiple variants — supporting React + Vite, Express APIs, MongoDB, Prisma, Firebase, and SQL setups.
- Scaffolding: Spins up production-ready project structures in seconds instead of hand-rolling boilerplate each time
- Variants: Mix and match frontend (React + Vite) and backend (Express) setups with your choice of database layer — MongoDB, Prisma, Firebase, or SQL
- Published as a package: Installable and usable directly via npm
GitHub: github.com/PuneethV333/Vima-Clt-tool Live: vima-clt-tool.vercel.app npm: npmjs.com/package/vima-clt
A full-stack co-living marketplace connecting tenants and property owners, built with semantic matching under the hood.
- Frontend: React/TypeScript, Firebase Auth, TanStack Query, Tailwind CSS
- Backend: Node.js, MongoDB, Redis, Qdrant (vector DB) with Google Gemini embeddings for semantic property and roommate matching
- Features: Multi-step onboarding for tenants and owners, property browsing and detail pages, a create-property wizard, a roommate preference survey, and an owner dashboard
GitHub: github.com/PuneethV333/Co-Living Live: co-living-two.vercel.app
Backend Intern — Unified Mentor Worked on the Co-Living platform above, focused on the MERN stack: building out backend services, API design, and data modeling for a real production-style application.
Backend
Frontend
Databases
AI / Infra
B.Tech in Computer Science Engineering — PES University, Bengaluru
Feel free to explore my repos or reach out — always happy to talk about backend architecture, agentic AI, or anything I'm building.