Gabi Dobocan

Wise Wombat is an advanced AI agent platform designed to drive lead generation, sales qualification, and content marketing across multiple channels, all through conversational interfaces.

• Built a scalable multi-LLM microservice architecture enabling dynamic prompt routing, retrieval-augmented generation (RAG), model context protocol (MCP), and robust job orchestration with validation pipelines.
• Integrated conversational outreach across platforms, including Telegram, Reddit (posts + real-time chat), automating lead engagement and qualification at scale.
• Automated content generation and publishing workflows for Hashnode, LinkedIn, and X (formerly Twitter), including image and audio assets, powered by multimodal models.
• Engineered a full-stack system with Node.js, tRPC & MongoDB services, a React-based admin dashboard, cross-platform Electron desktop clients, WebSocket-based proxy server, and Stripe payments.
• Implemented browser-based automation via Puppeteer for custom scraping, interaction, and validation workflows.
• Built remote-triggered, local Puppeteer automation controlled by a backend microservice over socket connections, enabling safe and distributed browser control.
• Scaled the system to support hundreds of users and tens of thousands of qualified leads per month.

Cozy Mantis is a marketplace for high-quality, production-grade custom nodes and workflow templates for ComfyUI - the most modular and extensible interface for diffusion-based image and video generation.

• Developed advanced ComfyUI nodes in Python, integrating support for character consistency, virtual try-on, animation, and multi-stage video generation.
• Leveraged cutting-edge image and video models, including Stable Diffusion 1.5, SDXL, and Flux for images, and SkyReels, LTXVideo, and Mochi for video workflows.
• Created AI-generated influencers and virtual personas that drove thousands in revenue, using fully automated pipelines for long-form content, ads, and trailers.
• Enabled rapid production of high-quality video assets at a fraction of traditional cost, drastically reducing the time and expertise required to produce complex visual media.

Sandworm is a modern, open-source toolchain for identifying and responding to software supply chain vulnerabilities in JavaScript and PHP ecosystems. It combines real-time threat detection, large-scale static analysis, and ML-powered intelligence.

• Developed the core open-source CLI tool in Node.js for scanning JavaScript and PHP projects (npm & Composer) for known vulnerabilities and suspicious patterns.
• Implemented Sandworm Guard, a runtime monitoring and alerting system that enforces policies and detects unexpected behavior during npm install, protecting developers from dependency-based attacks as they happen.
• Architected a cloud-scale scanning backend using AWS Lambda and MongoDB, indexing and analyzing all published npm and Composer packages in near real-time.
• Built the web-facing audit platform using Next.js, Eleventy, AWS S3, CloudFront, and Lambda, including frontend UI design, static site generation for audit reports, and Stripe payments.
• Applied LLM-assisted and ML-based techniques to uncover novel attacks and obfuscated malware, catching hundreds of active threats within minutes of deployment.
• Contributed to developer education and ecosystem awareness through technical blog posts, documentation, and threat analyses.