About me
I build AI tools and ship them to production. I handle prompts, retrieval, and the UI. I focus on reliability and keeping the interface easy to use. Experienced in LLM memory design, real‑time pipelines, browser automation, macOS apps, and SaaS.
Skills
Projects
Verse is a Chrome extension that turns Chrome into an agentic browser. It adds a chat sidebar where you can tell it what to do, like fill forms, summarize or navigate websites, or automate simple workflows. You can hand it tasks across multiple tabs and it keeps working quietly in the background while you continue browsing.
Local, multi‑agent AI web automation built for Manifest V3. Runs fully in the browser: a background service worker coordinates specialized agents that plan steps, act on pages, and validate outcomes. React + TypeScript UI for the side panel and settings. Fast, modular builds for quick iteration.
- Chrome MV3 service worker + content scripts
- React 18, TypeScript (strict), Tailwind CSS
- LangChain.js with multi‑provider models (OpenAI, Anthropic, etc.)
- pnpm workspaces + Turbo; Vite per workspace
- Typed storage over chrome.storage; zod-style schema utils
- ESLint/Prettier, Vitest; HMR for dev
- Planner: decomposes goals into actionable steps.
- Navigator: executes DOM actions via content scripts (click, type, select, scroll, navigate).
- Validator: verifies results and completion criteria.
- User prompts in the side panel.
- Planner produces a step plan; background sends commands via Chrome messaging.
- Content script performs page actions and returns DOM snapshots.
- Validator checks targets/states and signals next steps or completion.
I started by building a Chrome extension to solve form-based exams. I used it to complete over 50 assignments where I had to answer form-based questions. With one click, the agent would send the top unanswered question to an LLM API, get the answer, and fill it in automatically. It repeated this until all answers were filled.
One thought stuck with me: if it can solve these questions perfectly, what else can it do? Maybe everything on the web?
That's why I built Verse, inspired by Cursor (a coding agent), but for Chrome. Just ask it to do anything, and it does it by itself.
Luca is an AI copilot for online meetings. It listens to everything happening in the meeting, can see your screen, and answers questions about both the meeting and what's on your screen in real time. It also gives live recaps if you miss anything. After the meeting ends, Luca automatically creates clear notes, a full summary, and saves the transcript.
macOS app built with SwiftUI + AppKit for overlay windows. Multi-window system (control panel, chat, transcription) that stays accessible during meetings. Real-time audio transcription via Deepgram, screen capture with OCR, and AI-powered memory system for context-aware responses.
- Swift 5.9+, SwiftUI, AppKit
- ScreenCaptureKit (screen/audio capture)
- AVFoundation (audio processing, format conversion)
- Vision framework (OCR)
- Node.js backend (Vercel serverless)
- OpenAI API (GPT-4o-mini, embeddings, vision)
- Deepgram Nova-2 (real-time STT via WebSocket)
- Firebase (Auth, Firestore)
- Redis (auth state, rate limiting)
- Vector memory system (1536-dim embeddings)
It's inspired by Cluely. I really loved the idea that AI should have an interface that's easily accessible anywhere and blends into workflows.
While attending meetings and calls, it was hard to keep notes and write more stuff related to it. That's why I built a meeting assistant to help with that.
Zest is an event management and booking platform like BookMyShow. Artists, event organizers, and venues can list events as easily as posting on Instagram, then manage finances and attendees from a clean dashboard. Users can browse and book any listed event seamlessly from their phones.
Next.js 14 App Router app with server/client components. Domain-driven architecture for scalability. Session-centric event model supports multi-day events with multiple sessions. Real-time dashboard updates via Firebase listeners. Razorpay integration for payments. Built for the Indian market with focus on small events and local venues.
- Next.js 14.1.0 (App Router, SSR/SSG)
- React 18.2.0, TypeScript 5 (strict)
- Tailwind CSS 4.1.8, CSS Modules
- Firebase (Auth, Firestore)
- Firebase Admin SDK (server-side)
- Razorpay (payment gateway)
- Real-time listeners (onSnapshot)
- Atomic transactions (data consistency)
- Domain-driven design architecture
- Vercel deployment (serverless)
I wanted to reduce the hassle of waiting in lines and make everything easily accessible. Anyone who wants to enjoy should be able to book from their phone. Not just big concerts, but also small events and activity places, helping users get access and helping businesses reach their audience.
I believe offline is the new luxury. People will spend on moments and crave in-person connections even more.
Lens tracks how your brand appears in AI-generated responses (ChatGPT, Claude, Perplexity). It automatically extracts business information from your website, creates canonical descriptions, identifies relevant AI search topics, finds competitors, and simulates how AI assistants respond to queries about your space.
Next.js 14 App Router application with multi-step onboarding flow. Automatically scrapes websites to extract business information, generates AI-powered descriptions, identifies relevant topics and competitors, and simulates AI search responses to track brand visibility metrics.
- Next.js 14 (App Router)
- TypeScript, Tailwind CSS
- Zustand (state management)
- OpenAI GPT-4
- Lucide React (icons)
- Website scraping
- AI search simulation
- Brand visibility tracking
- Website Scraping: Automatically extracts business information from your website
- AI-Generated Description: Creates a canonical business description
- Topic Discovery: Identifies relevant AI search topics for your business
- Competitor Analysis: Finds and tracks your competitors
- AI Search Simulation: Simulates how AI assistants respond to queries about your space
- Visibility Metrics: Tracks brand visibility percentage, ranking, and share of voice
- Website Input → Enter company name and website URL
- Description → AI scrapes site and generates description (editable)
- Topics → Select 3+ topics you want to rank for in AI responses
- Competitors → Review and edit auto-suggested competitors
- Simulation → View AI search results and see where you appear
- Dashboard → Track visibility metrics over time
Education
BTech. AI & Data Science at K. J. Somaiya College of Engineering, Mumbai