Amit Kumar

MyKernel — 64-Bit x86_64 Operating System Kernel

A complete, educational, modular 64-bit operating system kernel built from scratch in C and NASM assembly for the x86_64 architecture. It boots into long mode via the Multiboot2 specification and is bootable through GRUB2.

Key Features

• Boot & CPU: Multiboot2-compliant bootloader, 64-bit long-mode trampoline, GDT, 64-bit IDT with all 256 vectors mapped, TSS, exception gate handlers for CPU faults (page fault, GPF, division by zero) and hardware IRQs.
• Memory management: a bitmap physical page allocator (PMM) at 4 KiB granularity, a 4-level paging virtual memory manager (PML4 → PDPT → PD → PT), and a kernel heap with kmalloc/kfree/krealloc supporting block splitting and merging.
• Processes & IPC: context switching, a thread scheduler, spinlocks, mutexes, inter-process message queues, ring-buffer pipes, and shared memory.
• Filesystem: a Virtual File System abstraction with a FAT32 partition reader/driver and a log-structured JournalFS.
• Networking: a complete zero-copy stack — Ethernet frames, an ARP table, IPv4, ICMP echo (ping), UDP sockets, a TCP state machine (SYN/ACK/ESTABLISHED), and a DHCP client for IP address leasing.
• Drivers: VGA text (80×25) and graphics output, serial COM1 UART (115200 baud), an interrupt-driven PS/2 keyboard with modifier-key handling, a PIT timer at a 100 Hz tick rate, RTC, and PC speaker.
• Power & security: ACPI power-off and reboot, SHA-256 and MD5 checksums, and access-control verifiers.
• Utilities: an interactive kernel shell CLI, a structured logger (kprintf), a stack-trace debugger, and a freestanding C standard library.

Challenges Solved

The project required implementing, from first principles and without an underlying OS, the full stack a kernel normally depends on: transitioning the CPU into 64-bit long mode, building paging and heap allocation without a libc, writing interrupt-safe drivers, and implementing a TCP/IP network stack down to the Ethernet frame level — all validated through boot testing in QEMU across Windows, macOS, and Linux build environments.

ThreatVision AI

An open-source, modular computer-vision framework for real-time threat detection, multi-object tracking, spatial analytics, and multi-channel incident response. It processes streaming video from live cameras, RTSP network streams, video files, or static images, and computes a calibrated threat score for human operator review rather than issuing absolute automated judgments.

Key Features

• A modular pipeline covering camera ingestion, object detection, multi-object tracking, behavior analysis, threat scoring, and human review, assembled from independently maintained detectors — person, vehicle, weapon-shape, fire/smoke, fall, and crowd density — combined behind a common interface.
• A transparent, configurable threat-scoring engine that converts per-frame detections into zone- and time-aware incident scores, using temporal persistence checks and zone rules to reduce false-positive alert fatigue.
• A web dashboard and command-center UI, plus a cloud REST API and WebSocket streaming interface for integration into external systems.
• Combines single-stage detectors (YOLO-family) for real-time streaming inference with two-stage detectors (Faster R-CNN) for higher-accuracy offline re-verification of borderline incidents.

Challenges Solved

The framework is designed as an operator-assistance tool rather than an autonomous decision-maker: detections return explicit confidence bounds and threat scores instead of absolute assertions, addressing alert fatigue, opaque decision logic, and vendor lock-in commonly found in closed, proprietary security appliances, while keeping every stage from detection to scoring inspectable and replaceable through a plugin system.

Searcher Browser

A modern, lightweight desktop web browser built from scratch in Python using the Chromium-based PyQt6-WebEngine rendering engine, packaged as a single portable executable requiring no installation.

Key Features

• A built-in AI Assistant that summarizes webpages, answers questions about the current page, generates study notes, and automatically groups tabs by topic.
• A native ad blocker integrated into the network interceptor, a password manager, and an incognito mode that retains no history, cookies, or cache.
• Offline mode with 500 MB of page caching, offline-aware address-bar search of local history and bookmarks, and a built-in offline mini-game.
• Productivity features including a custom start page, full history and bookmarks, session restore, a download manager, light/dark theming, voice search, and support for multiple search engines (Google, Bing, DuckDuckGo, and others).
• A mobile-view toggle for emulating responsive device environments during development and testing.

Challenges Solved

The project packages a full Chromium-based browsing engine, local SQLite-backed data storage, and an AI assistant sidebar into a single cross-platform, no-install executable via PyInstaller, targeting Windows, macOS, Linux, and mobile (as an installable PWA).

StepDistanceCalculator

A professional, object-oriented Python package that converts physical distances into biomechanically accurate walking step counts, offered through both a high-level Python API and a command-line interface.

Key Features

• Single- and multi-city/route step calculations, with exact and rounded step counts across multi-segment route chains.
• Gender-based default step lengths (adult male 0.78 m, adult female 0.70 m, neutral fallback 0.74 m) and biomechanical height-based estimation using validated height-to-step ratios, plus support for custom stride lengths.
• Activity pace modifiers for walking (1.00×), brisk walking (1.08×), jogging (1.25×), running (1.40×), and hilly/uphill terrain (0.90×).
• Automatic unit conversion across meters, kilometers, miles, feet, centimeters, and inches.
• A clean object-oriented architecture built around Person, Location, Segment, Route, DistanceCalculator, and ReportGenerator abstractions.
• Report export in plain text, JSON, CSV, and formatted PDF, plus Matplotlib visualizations — a bar chart of segment steps, a pie chart of percentage distance contribution, and a line chart of cumulative steps across checkpoints.
• An interactive terminal menu mode alongside the CLI, with complete input validation, a custom exception hierarchy, and structured logging.