SYS.INIT [0%]
Compiling interactive proof of work...

Not a Website. An Interactive Product.

Parth
Varekar

B.Tech Computer Engineering student (Mumbai University, 2024–2028) building local-first AI systems — speech pipelines, RAG knowledge bases, browser AI safety, and educational games. I ship to learn, and I learn by shipping.

Architectural Blueprint // 01

Who I Am.

I'm Parth — a B.Tech Computer Engineering student at K.C. College of Engineering (Mumbai University), class of 2028. I build AI systems because I'm curious about how intelligence can be engineered to run locally, privately, and reliably — not just routed through someone else's API.

In 2026 I completed a 120-hour Data Science & Analytics internship at Imarticus Learning (A+ grade), where I worked with SQL, Python/Colab, and Power BI on real datasets. Outside of coursework I ship projects — six of which are documented below, all verified and on GitHub.

Based in Mumbai. Open to internships and collaborations in AI systems, full-stack engineering, and developer tools.

SYSTEM.ID // 02

What I Build.

I build across the AI engineering stack — from C++-backed speech pipelines (whisper.cpp + llama.cpp) to in-browser Python execution (Pyodide/WebAssembly) to Chrome MV3 extensions that intercept and inspect AI agent traffic at runtime. Most of my projects are local-first: your data stays on your machine.

I care about architecture over hype. The projects below are real, documented, and reproducible — not API wrappers with a landing page.

SYSTEM.LOGIC // 03

How I Think.

Systems thinking over surface fixes. When a latency budget slips or a detection engine returns false positives, I don't patch the symptom — I re-examine the pipeline. Most of my projects started as personal problems I wanted solved properly: a GATE prep tool that didn't exist, a dictation daemon that respected my privacy, a safety layer for browser AI agents that nobody had shipped yet.

I believe in controlling the stack end-to-end — from the compiled C++ binary to the last pixel on screen. Theory is the starting point; shipping is the test.

SYSTEM.EDGE // 04

Where I Started.

My first shipped extension was Color Vision Assistant (2025) — built with a team to help low-vision users browse the web. It implemented a “partial blindness” mode via CSS/JS overlays that boosted contrast, font weight, and brightness. Crude by today's standards, but it's where I learned Chrome MV3, content scripts, and that software can quietly change someone's day. It's why accessibility still informs how I build.

Currently exploring:

  • _ Local LLM deployments & quantization (whisper.cpp, llama.cpp, Ollama).
  • _ In-browser Python execution via Pyodide/WebAssembly.
  • _ Multi-agent systems for media intelligence & content analysis.
  • _ Runtime safety layers for browser-based AI agents.

System.Nodes // Projects

Deployed
Architecture.

WhisperFlow preview — offline speech-to-text pipeline

AI VOICE // DEEP

WIP

WhisperFlow

Offline, zero-cloud speech-to-text + LLM pipeline. whisper.cpp transcribes, llama.cpp reasons — no internet after setup.

[ Python ] [ whisper.cpp ] [ llama.cpp ]
StudyOS preview — GATE prep PWA dashboard

FULL-STACK // DEEP

LIVE

StudyOS

A local-first PWA that runs my GATE 2027 prep like a SaaS product — 13 Prisma models, test-runner state machine, offline-first.

[ Next.js 16 ] [ Prisma ] [ PWA ]
Nexus-AI preview — educational coding game with in-browser Python

GAME + WASM // DEEP

WIP

Nexus-AI

A 2D sci-fi educational game where players write real Python in-browser via Pyodide/WebAssembly. Custom Canvas engine + level editor.

[ Pyodide ] [ Canvas ] [ CodeMirror 6 ]
2'nd_Brain preview — local RAG knowledge graph

RAG SYSTEM // DEEP

WIP

2'nd_Brain

Local-first RAG knowledge base. Dual-store (SQLite + ChromaDB), streaming SSE answers, knowledge-graph API, Playwright scraper.

[ FastAPI ] [ ChromaDB ] [ Gemini/Ollama ]
Agent Safety Net preview — Chrome extension for AI agent safety

BROWSER SAFETY // DEEP

WIP

Agent Safety Net

Chrome MV3 extension — runtime safety layer for browser AI agents. Intercepts fetch/XHR, detects PII + prompt injection at <1ms.

[ TypeScript ] [ Chrome MV3 ] [ React ]
Shorts Intelligence OS preview — multi-agent YouTube Shorts analysis

MULTI-AGENT // DEEP

WIP

Shorts Intelligence OS

Multi-agent CLI that analyzes YouTube Shorts — viral scoring, retention forecasts at 3s/10s/20s, scene scripts. 15 formally specified metrics.

[ FastAPI ] [ NVIDIA NIM ] [ FAISS ]

Topology // Tech Stack

System
Capabilities.

├── AI_SYSTEMS
├── whisper.cpp + llama.cpp (local STT + LLM)
├── RAG (ChromaDB, FAISS, pgvector)
└── Multi-agent pipelines (LangGraph, NVIDIA NIM)
├── BACKEND_INFRASTRUCTURE
├── Python, FastAPI, Prisma, SQLite
├── SSE streaming, Pyodide/WASM, Playwright
└── MySQL, Power BI, REST APIs
└── FRONTEND_UX
├── Next.js 16, React 19, TypeScript, Tailwind
├── shadcn/ui, Canvas API, CodeMirror 6
└── GSAP, Chrome MV3, PWA / Service Workers

System.Logs // History

Version
Control.

v3.0.0 [Current] -> AI Systems Builder
@ Mumbai // 2026 — Present
+ WhisperFlow: offline STT + LLM pipeline (whisper.cpp + llama.cpp)
+ Agent Safety Net: Chrome MV3 runtime safety for browser AI agents
+ StudyOS: local-first GATE prep PWA (Next.js 16 + Prisma)
v2.0.0 -> Data Science Intern // Imarticus Learning
@ Mumbai // Feb — Mar 2026 // Grade: A+
+ 120-hour internship: MySQL, Python/Colab, Power BI
+ Built interactive dashboards on real datasets
v1.0.0 -> B.Tech CE begins + First Extension
@ K.C. College, Mumbai University // 2024 — 2025
+ Admitted via MHT-CET to B.Tech Computer Engineering
+ Shipped Color Vision Assistant (Chrome MV3, team project)
+ Full-Stack Java certification (EduSkills, A+ grade)

PIPELINE.VISUALIZER // How Data Flows

Watch the
Pipelines Run.

Animated data-flow diagrams for all six of my projects. Particles represent data moving through each pipeline stage. Switch between projects to see how architecture differs.

WhisperFlow — Audio → ffmpeg → whisper.cpp (STT) → llama-server (LLM) → Win32 SendInput (text injection). Fully offline, zero cloud calls.

Global Routing
Interface.

Execute commands to traverse the portfolio system, open project telemetry, or initiate contact protocols directly.

> Try command: projects

> Try command: open whisperflow

> Try command: contact

05 // CONVERSION PROTOCOL

Initiate
Handshake.

Transmit Email
[ GitHub ] [ LinkedIn ] [ +91 7400082627 ]
BG
SFX