Overview

Quiescan is a static malware analysis engine that never executes a file — it extracts, analyzes, and scores everything from the binary itself. It combines 500+ YARA rules with a 10M+ malware hash database and Google Gemini AI to produce a structured verdict with threat level, matched indicators, and a plain-English explanation.

Context

The goal was to make malware triage accessible to anyone — not just analysts with sandboxes and reverse engineering expertise. Upload a suspicious file, get back extracted strings, entropy scores, file hashes checked against the database, YARA matches, and an AI-generated threat summary in seconds.

The backend runs on Railway (FastAPI/Python) with Firestore for hash lookups and rate limiting. Anonymous users get 1 scan/day via cookie + IP hash; authenticated users get up to 50MB and 10 scans/day. A Gemini model fallback handles upstream 503s gracefully.

Tech Stack

Python

Firebase

Vercel

Overview

VOIDSTRIKE is an AI-powered recon suite that wraps 27+ security tools — nmap, nikto, sqlmap, and more — behind a single intelligent interface. Instead of running tools blindly, VOIDSTRIKE fingerprints the target first, then builds a custom scan plan and aggregates results into a single ranked report.

Context

Bug bounty hunters and pentesters spend a disproportionate amount of time chaining tools together and parsing their separate outputs. VOIDSTRIKE collapses that workflow into one command. The AI layer decides what to scan based on the target's profile, prioritizes findings by exploitability, and surfaces the highest-impact issues at the top of the report.

Tech Stack

Python

React

Docker

Shell

Overview

PITCHFORGE takes an app store URL or a plain-text brief and generates launch-ready marketing copy for App Store, Google Play, Twitter/X, Instagram, LinkedIn, and Product Hunt simultaneously. It includes ASO scoring, A/B variant generation with an AI critic, a 10-check launch readiness checklist, an iOS keyword field packer (greedy knapsack, 100-char budget), and public campaign microsites.

Context

Built for HackOnVibe 2026. The core problem: indie devs and small teams ship great apps but lose visibility because platform-specific marketing copy is tedious to write well. PITCHFORGE automates the entire launch content pipeline in one shot — from ASO keywords to LinkedIn announcements — while an AI critic evaluates each variant and suggests improvements.

The stack runs React 18 + Vite 7 on the frontend with a Node/Express + tRPC + Drizzle ORM backend backed by TiDB Serverless. Auth via Clerk. AI via Google Gemini. Deployed on Railway with Cloudflare Workers AI for edge inference tasks.

Tech Stack

React 18

TypeScript

Vite

Node.js

tRPC

Drizzle ORM

TiDB

Clerk

Gemini

Overview

VOIDWEN merges two agent personalities — a "caveman" (ultra-compressed prose) and a "ponytail" (YAGNI minimalist) — and adds a wenyan (classical Chinese) output mode via a fine-tuned ONNX model running entirely in the browser via transformers.js. The result is an AI that communicates with extreme density while preserving technical accuracy.

Context

The key architectural principle: CJK characters only appear in runtime output and are detected via Unicode codepoint arithmetic — never as character literals in source. Zero recurring cost, zero server, zero API tokens for non-reasoning tasks. Deployed via GitHub Pages and Cloudflare free tier.

Tech Stack

TypeScript

ONNX

transformers.js

Overview

VoidState tracks how you use your phone and PC, runs local ML models to find behavioral patterns, and surfaces insights in your choice of 7 tone modes — including Hinglish and an "Abusive Friend" mode that tells you the truth without filters. No data ever leaves your device.

Context

The differentiator: VoidState analyzes your own private behavioral data automatically rather than requiring you to manually log anything into a chatbot. The Android app runs TFLite models locally; the Windows desktop app uses ONNX via Tauri v2 + React. SQLite stores everything on-device. Clerk handles optional multi-device identity with Supabase E2E sync.

Tech Stack

Kotlin

TFLite

ONNX

Tauri v2

React

TypeScript

SQLite

Overview

CS-AI is a full-stack AI chat application deployed on Vercel with Firebase as the backend. It supports multiple AI models, Google and email authentication, per-user token quotas enforced server-side, persistent chat history, and a full admin panel for managing users and tiers.

Context

I wanted to build something that felt like a real product, not just a demo. The core challenge was making multi-model AI accessible while keeping it controlled. Different users get different access levels and every request is authenticated and rate-limited before it ever reaches an upstream API.

I built the entire stack myself: the chat UI, the serverless API layer that proxies Gemini and OpenRouter, the Firestore schema, the token quota system, and the admin panel. Streaming responses were handled via Server-Sent Events with a filter that strips internal model reasoning before it reaches the user. Rate limiting runs on Upstash Redis at 20 requests per minute per IP.

Tech Stack

JavaScript

Firebase

Vercel

Gemini

OpenRouter

Overview

Notes AI is a web app that turns any topic, pasted text, or uploaded PDF into clean structured study notes using Google Gemini. It supports four note styles — Structured, Bullet Points, Cornell Method, and Mind Map — and lets you export everything as a formatted PDF or Markdown file in one click.

Context

The problem I kept seeing was students drowning in messy, unstructured notes that were hard to revise from. I wanted to build something that took whatever you had — a topic, a wall of text, a lecture PDF — and gave you something actually usable in under a minute.

The PDF export runs entirely in the browser using jsPDF with dark-themed pages, proper heading hierarchy, and accent-colored markers. The API key is stored in .env only — it never touches any server other than Google's.

Tech Stack

JavaScript

Vercel

Gemini

jsPDF

Overview

Student Data Administration System is a desktop application built entirely in Java using AWT and Swing, backed by a MySQL database hosted on Clever Cloud. It provides a clean role-based interface for school and institute administrators to manage student records with access restrictions enforced at the class level.

Context

The goal was to build something that worked like a real internal tool, not a prototype. Authentication uses BCrypt hashing so passwords are never stored in plain text. The dashboard renders all student records in a scrollable, sortable table with live search across name, class, roll number, and mobile.

Student management is gated behind the Admin role with confirmation dialogs to prevent accidental deletions. The Admin Panel lets admins create new user accounts, change roles, reset passwords, delete users, and configure per-user class access restrictions. Built and validated against 150+ official CWIT student records.

Tech Stack

Java

MySQL

Swing/AWT

Overview

CS MART is a developer-focused hardware storefront with a bold editorial design — stark black backgrounds, acid-green accents, and oversized headline typography ("Gear up. Code more."). The catalog covers curated keyboards, desk gear, and developer essentials with a clean product browsing experience including search, bag, and field notes sections.

Context

The design direction is intentionally anti-Amazon — no clutter, no noise. Product cards are stripped down, the hero section leads with one product at a time (OBJ / 001, Field Series), and every detail communicates craft over commerce. The acid-green CTA and CS logo lock the brand into a distinct identity that leans developer-subculture over generic retail.

Built as a frontend-only project to push CSS layout and typographic hierarchy without a framework. Live at csmart.chinmayshinde.tech.

Tech Stack

Overview

STEALTHWALL is a self-hosted ML-driven IPS and WAF that runs as in-process middleware or a reverse-proxy sidecar. It tracks requests per source IP in 60-second sliding windows, computes 14 normalized statistical features (Shannon entropy, inter-arrival variance, status code distribution, enumeration entropy, signature presence), and runs an optimized LightGBM model via ONNX Runtime in sub-millisecond inference time.

Detected threats trigger graduated response tiers: rate limiting, proof-of-work challenge, provisional block, temp block, long cooldown. Shared-IP protection handles CGNAT and proxy gateways safely. A real-time dark-theme Operations Console (WebSocket feed + Prometheus metrics) ships alongside the core.

Architecture

Four-stage pipeline — Pre-Enforcement Gate (iptables/nftables cooldown check), Feature Extraction (sliding-window stats), ML Inference (LightGBM ONNX with pure-Python fallback), Graduated Response. Works as FastAPI/Starlette middleware (Python) or Express middleware (Node.js). Drop-in integrations also exist for Next.js Edge Middleware, Nginx, and PHP/WordPress.

Published to both PyPI (pip install stealthwall) and npm (npm install stealthwall). Full Docker Compose stack bundles the dashboard, Redis, and Prometheus. CLI includes a built-in attack simulator to test defenses against sqlmap, directory bruteforce, and exploit probes.

Tech Stack

Python

LightGBM

ONNX Runtime

FastAPI

Node.js

Redis

Prometheus

Docker