AI • Automation • Full-Stack Development
Computer Science / Data Science student who builds real software, experiments with AI and automation, and learns by shipping.
I build AI-powered software, automation systems, full-stack applications and experimental products. I learn by building, and I share what I make publicly, from hackathon prototypes to agent systems and mobile apps.
Current focus: AI agents, browser agents, on-device AI, and automation workflows.
|
A privacy-preserving browser agent with on-device screen understanding. PAROKSH is a working browser agent I built. It understands what is on screen and interacts with the browser on the user's behalf, and it includes a privacy shield. Screen understanding happens on-device.
|
| AI Agents Browser agents, tool-using assistants |
AI Applications Fact-checking, language learning, medical avatars |
Automation n8n workflows, API integrations |
| Full-Stack Systems React, Node.js, Express, real-time updates |
Mobile Apps Flutter and Dart |
Developer Tools Pipelines, downloaders, experiments |
A personal assistant built with n8n that connects to Gmail, Google Calendar, Docs, Sheets, web research, Reddit, Hacker News, GitHub, weather, maps, web scraping, monitoring (job postings, price changes), Excel export and Telegram. It has a custom website interface with typed and voice input using browser speech APIs.
I started with a multi-agent architecture, but 60-90 second responses and agent loops pushed me to a flatter design, which brought responses down to single-digit seconds. I also experimented with self-hosted Ollama models.
n8n AI Agents Ollama Telegram Browser Speech APIs
A language-learning AI platform built by Team CodeWithMay for the Language track of the Adaption AutoScientist Challenge x HackIndia. It includes a voice coach with speech recognition and text-to-speech, 12 languages, 9 conversation modes (interview prep, IELTS, business, debate), grammar checking, translation with confidence scoring, AI vocabulary lookup, a provider abstraction layer with local-model / AutoScientist model switching, and a dataset export pipeline.
A real-time medical conversational avatar platform built for HackIndia 2026. It supports English, Hindi and Hinglish, with emotion detection, intent classification and multiple AI personas. I generated synthetic patient-AI conversations and built a dataset of 1,018+ real + synthetic medical dialogues with Adaption, exported to Kaggle.
A Flutter app that verifies text and images (with OCR) against multiple external sources, including Google Fact Check, Wikipedia, news APIs, WHO, UN and PIB. It produces verdicts, confidence scores, source breakdowns and explanations.
Flutter Node.js Express GPT OSS 120B LLaMA 3.1 OCR
A full-stack media downloader supporting platforms such as YouTube, TikTok, Instagram and Pinterest. It shows real-time download progress over Server-Sent Events, handles background processing, and cleans up temporary files. The UI is platform-aware.
React Tailwind CSS Node.js Express yt-dlp SSE
More projects
Daily Focus: a Flutter productivity app with daily task tracking, persistent notifications, automatic midnight failure status for incomplete tasks, local history and background processing. Built with a three-layer Clean Architecture (Domain, Data, Presentation) using Riverpod, Hive, flutter_local_notifications, flutter_background_service and WorkManager.
Instagram Reel Automation: an n8n workflow that reads video URLs and captions from Google Sheets, fetches videos from Cloudinary, uploads Reels through the Meta Graph API, waits for processing, publishes, and updates the sheet. Covers REST APIs, OAuth, webhooks, asynchronous workflows and error handling.
YouTube Visual Summary Pipeline: a Google Opal workflow that takes a YouTube URL, extracts and cleans the transcript, pulls out concepts, statistics, arguments and takeaways, and turns them into visual cards on a generated HTML summary page.
| Event | Result |
|---|---|
| OpenClaw Hackathon 2026 (HackIndia) | First Runner-up among 200+ teams, for an AI system for real-time misinformation detection |
| Bharatiya Antariksh Hackathon 2026 | Participant (presented by ISRO, powered by Hack2skill) |
| HackDavengers 1.0 | Participant in the 8-hour online hackathon organized by Devengers |
| Google Cloud Arcade Facilitator Program 2026 | Enrolled |
| AI & IoT Vocational Training | Completed a 30-day program at CSVTU-FORTE, Bhilai |
Languages
Frontend and Mobile
Backend
AI and Automation
LLMs • AI agents • Vision-Language Models • Computer Vision • OCR • Synthetic data • On-device AI
Tools and Concepts
REST APIs • OAuth • Webhooks • Server-Sent Events
