Building software from interface to infrastructure.
I'm Aashutosh Joshi, founder of XRGO Securities. I engineer production systems — autonomous AI pipelines, low-latency distributed backends, and defensive cyber platforms.
Production systems built for scale & security.
A portfolio of real-world software across offensive security, autonomous AI agents, reverse-engineered protocols, collaborative streaming, and self-hosted fintech systems.
XRGO Securities
An enterprise offensive security platform designed for continuous external attack surface discovery, automated port auditing, CVE detection, and cloud perimeter hardening. Automates hours of manual pen-testing routines into reproducible sub-minute assessment pipelines.
- ›Automated reconnaissance engine integrating custom scanner workers with Nmap and nuclei CVE signature databases
- ›High-speed async backend with PostgreSQL relational stores and distributed Redis queuing
- ›Real-time network security scoring, SSL/TLS posture auditing, and zero-trust Cloudflare edge policy enforcement
- ›Comprehensive automated vulnerability report generation for engineering teams and infrastructure operators
Python
Docker
Cloudflare
Offensive Security
PostgreSQL
Linux
Redis
JARVIS — AI Operating System & Agent
Conversational AI operating system that interacts directly with your machine. Executes real terminal commands, inspects Docker containers, navigates filesystems, audits processes, and automates workflows through a chat interface powered by LangGraph state machines and multi-provider LLM failover.
- ›Conversational interface with 59 integrated system tools across terminal execution, file management, and Docker control
- ›Multi-provider LLM failover engine dynamically routing requests across Groq Llama-3, Gemini Flash, and OpenAI
- ›Real-time interactive terminal streaming via WebSockets with risk assessment levels and human-in-the-loop approvals
- ›LangGraph cyclic state machines with SQLite checkpointed conversational memory and JWT + 2FA TOTP security
LangGraphPythonDocker
WebSocketsLinux
SQLite
Rudhra — Collaborative Streaming Platform
Collaborative movie and series watch-together platform built with Next.js 16, React 19, and Node.js. Features sub-frame synchronized playback via low-latency WebSockets, host-controlled playback authority, multipart episode uploads, and instant room generation.
- ›Sub-frame multi-user video playback synchronization with host authority over play, pause, and seek events
- ›Instant room creation with shareable session links and multipart movie & episodic media uploads
- ›High-throughput WebSocket broadcast server maintaining synchronized playback drift under 14ms
- ›Netflix-inspired dark interface built with Next.js 16 App Router, React 19, and Tailwind CSS
Next.js 16
React 19
TypeScript
Node.jsWebSockets
Tailwind CSS
Reverse-Engineered AI Routing Gateway
Reverse-engineered proprietary model endpoints and inference protocols (routing unreleased and preview models including gemini-robotics-er-2-preview, gpt-5.0-luna, and gemini-3.5-flash-lite) to bypass restrictive limits and process over 133,998,102+ tokens through a custom, resilient gateway with live telemetry.
- ›Reverse-engineered proprietary model endpoints, session tokens, and streaming RPC protocols
- ›Routed 133,998,102+ total tokens with automated multi-model traffic distribution and failure recovery
- ›Engineered live telemetry dashboard tracking 7-day token surges, model cost accrual, and active vision routing
PythonReverse EngineeringDockerRedisLinux
Auto-Clipper Vision Studio
Autonomous computer vision and acoustic intelligence engine that ingests long-form video streams, detects high-virality narrative moments, and automatically renders 9:16 vertical video clips with dynamic face tracking and animated karaoke subtitles.
- ›Neural speaker diarization with Pyannote to isolate distinct audio speakers and identify narrative climax hooks
- ›Real-time face detection and active speaker tracking via YOLOv8 and MediaPipe for smooth dynamic 9:16 vertical reframing
- ›Timestamped word-level speech transcription and acoustic alignment powered by Whisper deep learning models
- ›Autonomous Google Colab A100 GPU bridge daemon to offload high-resolution 60 FPS video transcoding and rendering
Python
OpenCV / YOLO
WhisperPyTorch
FFmpegReact 19
Lumor Pay — Invoicing Platform
Comprehensive self-hosted invoicing and client payment platform engineered with React, TypeScript, Node.js, and SQLite. Features dual-gateway checkout (Creem.io & PayPal), automated transactional SMTP invoice delivery, payment simulation testing, and auto-clearing port management.
- ›Self-hosted client invoicing engine with dynamic invoice lifecycle and real-time revenue analytics
- ›Dual payment gateway integration supporting Creem.io webhooks and PayPal checkout with simulation mode
- ›Centralized SMTP transactional mail dispatch for invoices and receipts with zero external cloud lock-in
- ›Zero-configuration transactional SQLite persistence with automated crash recovery and port recycling daemons
TypeScriptReactNode.js
ExpressTailwind CSSSQLite
AutoApply Agentic Engine
Autonomous evaluation and application engine that parses complex engineering role specifications, computes multi-dimensional candidate compatibility against ground-truth portfolios, and executes verified end-to-end submissions.
- ›Multi-dimensional semantic scoring comparing candidate background vectors against role requirements with zero hallucinations
- ›Deterministic question-answering engine strictly grounded on verified factual candidate profiles
- ›Automated browser interaction workflows handling multi-step application forms, file attachments, and validation barriers
PythonPlaywrightPostgreSQLRedisDocker
AgentTube Content Engine
Autonomous video production pipeline that analyzes trending topics, synthesizes structured narratives, coordinates neural voiceover generation, dynamically composites video footage via FFmpeg, and schedules multi-platform publishing.
- ›Fully unattended lifecycle from trend identification to rendered high-definition video assembly
- ›Dynamic multi-track video compositor synchronizing visual cuts, voiceover narration, and kinetic subtitle typography
- ›Containerized background rendering daemon running automated production schedules on hardened Linux server
PythonNode.jsFFmpegDockerLinuxTechnologies in production.
40 vetted languages, runtimes, distributed stores, and AI models utilized across production infrastructure.
Python
CORE AI & SECURITYFastAPI, Celery, LangGraph, computer vision & offensive scripts
Engineering systems with craftsmanship & precision.
I'm Aashutosh Joshi, a software engineer and the founder of XRGO Securities based in Kathmandu, Nepal.
My background spans the entire software spectrum: designing responsive frontend interfaces in React 19 and Next.js, building asynchronous backend microservices with FastAPI and Node.js, and hardening Linux infrastructure with automated penetration testing tools.
I founded XRGO Securities to solve real pain points in cloud security — turning hours of repetitive vulnerability assessments into autonomous, reproducible sub-minute scanning engines.
Speed as a Feature
Software should feel immediate. I prioritize sub-15ms server responses, aggressive in-memory caching with Redis, optimistic client state updates, and lean bundle sizes.
Pragmatic AI Systems
I build practical agentic workflows with LangGraph and computer vision with OpenCV/YOLO that solve real problems, rather than slapping generic LLM wrappers on everything.
Security by Default
As founder of XRGO Securities, defense-in-depth is in my DNA. Every endpoint is audited, edge perimeters are locked down with zero-trust policies, and all user input is sanitized.