LinkedIn Outreach Automation

Built a 24/7 Dockerized LinkedIn automation system with SQLite persistence and VNC/noVNC live monitoring. Implements stealth browsing, targeted search queues, personalized connection requests, follow-ups, and automated invitation cleanup. Includes dynamic templating ({{name}}, {{company}}) and configurable human-like delays with safety limits. Enables real-time visual observation even in headless execution.
Tech Stack: Node.js, Docker, React, TypeScript, SQLite, Puppeteer Stealth, VNC/noVNC, Xvfb
Impact: Fully autonomous outreach system for scalable networking automation. Reduced manual outreach workload by approximately 2–3 hours per day through continuous automated execution.

Implemented biometric authentication (WebAuthn) for a production web platform (Ad Opus), enabling secure passwordless login flows.
Tech Stack: React (TypeScript), GraphQL, Hasura, MobX-State-Tree, Tada GenQL, proprietary libraries.
Collaborated with UI/UX using Figma and managed development tasks via Jira and Bitbucket in an agile workflow.
Maintained and extended internal shared libraries, improving code reuse, system capabilities, and overall security.
Designed and implemented a desktop document scanning module as a Web Component within a microfrontend architecture (React frontend, Node.js + Express backend), replacing a commercial Scanbot SDK solution (~€7,200/year cost).
Built a full image processing pipeline: grayscale, contrast enhancement, sharpening, and binarization.
Developed a custom Python ML model for document binarization, improving output quality compared to standard OpenCV approaches and increasing OCR readiness.
Implemented automatic document edge detection with adjustable masking for precise user-controlled cropping.
Enabled export of processed documents into downloadable PDFs with optional OCR integration.
Result: Eliminated dependency on third-party scanning software and reduced operational costs by ~€7,200/year while maintaining high-quality document processing.

Developed an AI-powered customer support platform integrating automated assistance with human operator escalation, improving handling efficiency of support workflows.

Implemented NLP-based intent detection, sentiment analysis, and content moderation to classify user queries and determine urgency and escalation needs.

Built a multilingual response system supporting Romanian, Russian, and English, using a structured JSON knowledge base for scalable content expansion.

Designed and implemented a real-time chat system using WebSockets, enabling persistent sessions, message queuing, and automatic reconnection handling.

Implemented seamless AI-to-human handoff mechanism, passing full conversation context to operators for complex or sensitive cases.

Improved operational efficiency by reducing manual support load through automated resolution of routine customer inquiries while maintaining escalation coverage for high-complexity cases.

Contributed to the development of easyreser.io, an ERP system for the service industry, enabling efficient
management of operations. Developed statistical components using React.js and TypeScript. Created
interactive charts and dashboards to visualize data. Used Git and Bitbucket for version control and code
management. Collaborated closely with backend and design teams for seamless integration. Enhanced
user experience through intuitive interfaces. Implemented designs from Figma to ensure consistency and
efficiency in user interface development.

Built a deepfake image detection system using a custom CNN model trained on a Kaggle
dataset. The system classified manipulated images based on facial features and was extended into a
working Chrome Extension that processed YouTube video streams in real time, sending frames to a
backend API and returning a deepfake probability score.

Developed a full-stack Kanban task management application with user/admin functionalities. Utilized
React, TypeScript, Git, MySQL, and Express. Implemented features such as project management, user
authentication, task creation within projects, task assignment to registered users, task editing, status
updates via drag-and-drop functionality, and integration of a text editor for descriptions and comments.