I build practical software systems that connect AI, applications, and hardware — from internship work on embedded/IoT systems to full-stack products and local AI agents.
- 🤖 Local AI Agents with LangGraph, Ollama, Qwen, tool calling, validation, and human-in-the-loop workflows
- 🔌 Edge AI & Embedded Systems with Rock 5T/RK3588, XMOS audio hardware, Linux audio, I2C, I2S, and ALSA
- 🌐 Full-Stack Applications using React, Node.js, Express, Flutter, Spring Boot, MongoDB, and REST APIs
- 🛡️ Reliable AI Workflows with confidence checks, static validation, controlled execution, and offline fallbacks
- ☁️ Deployment & Engineering with GitHub Actions, Vercel, Render, testing, and production-style project structure
A Flutter-based IoT control application developed during my internship for discovering and managing relay hubs and device outputs through network communication.
Highlights: Flutter • UDP Discovery • REST APIs • PWM Control • Analog Monitoring • Scene Management • Persistent Settings
An embedded offline voice assistant developed during my internship and running on Radxa Rock-5T (RK3588) with an XMOS XVF3800 microphone array, local speech recognition, local LLM inference, and speech synthesis.
Pipeline: OpenWakeWord → PyAudio/ALSA → Whisper.cpp → Gemma via Ollama → Piper
Highlights: Embedded Linux • Edge AI • I2S/I2C • ALSA • Hardware Debugging • Local LLM • Engineering Handover Documentation
A full-stack platform I built to help students create, customize, publish, and track professional portfolios through a web-based builder.
Frontend: React 19 • Vite • React Router
Backend: Node.js • Express • MongoDB • JWT • GitHub OAuth • Cloudinary • Razorpay
Deployment: Vercel • Render
Highlights: Full-stack architecture • Authentication • GitHub OAuth • Dynamic portfolio URLs • Cloud media pipeline • Payments • Portfolio analytics
A privacy-first local developer agent that reviews Git changes, generates targeted pytest tests, validates AI-generated Python with AST checks, and waits for human approval before execution.
Highlights: LangGraph StateGraph • Ollama • Qwen2.5:3b • Python AST • pytest • Human-in-the-Loop • Controlled Subprocess Execution
A local-first agentic IT helpdesk that routes technical issues, executes tools, checks retrieval confidence, escalates uncertain cases into support tickets, and keeps working through a deterministic offline fallback.
Highlights: LangGraph • Ollama • Qwen2.5:3b • MCP • Streamlit • SQLite • Rich CLI • 115 automated tests • GitHub Actions
A role-based campus cleaning and verification system built for my college department to digitize daily assignments, cleaning completion, faculty verification, rejection handling, and reporting.
Highlights: Flutter • Spring Boot • MongoDB • REST API • Role-Based Workflow • Duplicate-Assignment Protection • PDF Reporting • Android APK
Python Java Dart JavaScript
LangGraph Ollama Qwen MCP Python AST pytest
React Vite Node.js Express Flutter Spring Boot
MongoDB MongoDB Atlas SQLite REST APIs JWT
Rock 5T RK3588 XMOS XVF3800 ALSA I2C I2S Embedded Linux
Git GitHub Actions Ruff Vercel Render
I am especially interested in systems where software has to do more than generate text:
- Stateful agent workflows instead of simple chatbot pipelines
- AI-generated output that is validated before execution
- Human approval and recovery paths for imperfect model output
- Local/offline AI for privacy and resource efficiency
- AI systems that interact with real hardware and operating-system services
- Full-stack products with clear architecture, testing, and deployment
My repositories showcase a progression from internship-based embedded and IoT development to full-stack application engineering, followed by local AI and agentic systems.
Some repositories contain internship-related integration work. Proprietary firmware, confidential source code, and private infrastructure are intentionally excluded.