CAPSULE
A TypeScript monorepo for building context-aware LLM applications with reusable packages and a browser extension.
AI ENGINEER / DEVELOPER
PUNE, INDIA
B.E. MECHANICAL ENGINEERING
AISSMS COE / 2028
Designing and building intelligent systems that solve real-world problems.
Machine Learning
Generative AI
AI Systems
Software Engineering
I am an engineering student exploring the intersection of artificial intelligence, software engineering, and real-world systems.
My interests lie in understanding how intelligent systems work, how they can be orchestrated, and how complex technical ideas become practical products. I believe in building, experimenting, questioning assumptions, and learning through implementation.
Programming, software development, and engineering fundamentals.
Early software workWeb development, automation, AI tools, and independent projects.
Full-stack and tooling experimentsMachine learning, neural networks, LLMs, agents, and system orchestration.
ISRO, Jarvis, Capsule, and systems workLanguages visible across the audited repositories.
Used in the space-weather and intercept-prediction work.
Local inference, voice systems, and model-serving experiments.
Explored through Jarvis and Capsule architecture.
Visible in the audited full-stack and API projects.
Build, testing, packaging, and repository workflows.
Seven projects. Different surfaces. One question: how does an idea become a working system?
A TypeScript monorepo for building context-aware LLM applications with reusable packages and a browser extension.
An AI-assisted component-search platform combining a Next.js application, Prisma data layer, Typesense search, and a Python AI worker.
An end-to-end ML pipeline for solar electron-flux forecasting with multiple model families and forecast horizons.
A Python desktop media toolkit with a PySide6 interface, FastAPI backend, FFmpeg processing, and an extension package format.
A local Python voice assistant organized around intent routing, memory, plugins, speech input, local inference, and tool execution.
A multi-view visualization application for project relationships, research loops, contradictions, and cognitive patterns.
An ESP32 medication-reminder concept with modular firmware, cup-weight verification, local storage, cloud sync, and a caregiver dashboard.
A modern web platform for engineering precision with CAD integration, built with React, Vite, and Three.js for 3D visualization.
A comprehensive healthcare platform for Asian Eye Hospital with appointment booking, doctor profiles, blog, and Sanity CMS integration.
A modern Indian fashion retail homepage concept with React, Vite, and Tailwind CSS featuring product showcase, cart, and responsive design.
A premium brand website for Jay Shankar clothing from Kolhapur with immersive hero, collections, and storytelling-focused design.
A Next.js dashboard for the ISRO space weather hackathon with Plotly.js visualizations, real-time solar data display, and interactive charts.
A comprehensive drone mapping and site intelligence platform with 3D visualization, interactive maps, machine tracking, and CAD integration demos.
A WebGL solar system controlled by hand gestures via MediaPipe — dual-hand particle field manipulation, planet focus, and orbital mechanics simulation.
A high-fidelity marketing website concept for VALORANT Mobile with interactive agent showcase, soundtrack terminal, pre-registration flow, and tactical HUD aesthetic.
A playful Valentine's Day interactive hub with multiple mini-experiences: question game, mini-game, and a "will you be my valentine" flow — all in a single-page navigation.
A React-based e-commerce frontend for devotional products with category browsing, product cards, and cart functionality.
Enterprise-grade IoT environmental monitoring platform with real-time sensor telemetry, EPA/ASHRAE threshold alerts, WebSocket notifications, and mathematical sensor simulation.
Modern full-stack media platform with React 19, Vite 7, Express, Supabase, Radix UI, and Framer Motion — comprehensive component library with authentication and analytics.
Local integration bridge connecting ChatGPT Web with Hermes Agent CLI through a Chromium extension and authenticated Python gateway — reduces manual copy/paste while retaining Hermes security.
Automated market news digest delivered to Telegram 3× daily (10 AM, 12 PM, 10 PM IST) on weekdays. Fetches 40+ RSS feeds, summarizes via Nvidia/Hermes (moonshotai/kimi-k3), runs free on GitHub Actions.
Original intelligent voice-controlled assistant with real-time speech recognition (Whisper), AI responses (Ollama Llama 3.1), command execution, and text-to-speech feedback.
Universal 3D gesture control library with MediaPipe Hands + Three.js. Zero runtime dependencies, ~18KB gzipped. Provides pinch, grab, point, swipe gestures for any Three.js scene.
Local-first encrypted vault for secrets, API keys, and credentials with client-side encryption, zero-knowledge architecture, and browser extension integration.
Multi-tenant drone mapping SaaS with project workspaces, orthomosaic processing pipelines, volume analytics, and client-facing dashboards.
Curated collection of production-ready ML models: time-series forecasting, classification, regression — with unified API, Docker serving, and benchmark suite.
High-conversion landing page for CreoVault CAD platform — hero with 3D viewer, feature grid, testimonial carousel, and waitlist capture with email automation.
Feature-rich YouTube downloader with yt-dlp backend, format selection, sponsor block removal, metadata embedding, and batch processing — GUI and CLI.
OBJECTIVE
How can context operations become reusable system primitives?
APPROACH
Token budgeting, semantic deduplication, ranking, provider boundaries, and package composition.
OBJECTIVE
How should physical signals become features for multi-horizon prediction?
APPROACH
Solar electron-flux data, physics-informed features, classical ensembles, and neural models.
OBJECTIVE
What modules are required between speech input and a safe system action?
APPROACH
Whisper transcription, Ollama inference, memory, plugins, executor boundaries, and speech output.
OBJECTIVE
How can software signals reflect a physical-world state?
APPROACH
ESP32 firmware, sensor modules, local storage, cloud sync, and caregiver-facing interfaces.