Furkan – Python, Typescript, Node.js

A backend for an IoT sensor monitoring platform that processes and analyzes large volumes of real-time data from distributed devices. The system handles high-throughput data ingestion, time-series storage, and asynchronous processing for alerts and notifications. It supports real-time updates for connected clients and delivers notifications through multiple channels. A separate reporting service generates structured reports in PDF and spreadsheet formats. The platform is containerized and includes a web-based dashboard with interactive data visualizations for monitoring sensor activity and system status.

• Designed and implemented the backend for a real-time IoT monitoring platform processing data from LoRa-based sensors;
• Built the data ingestion pipeline to consume MQTT messages via a message broker, normalize readings, and persist time-series data;
• Implemented real-time data delivery to clients using WebSockets;
• Developed an alerting system with configurable thresholds, cooldown logic, and multi-channel notifications;
• Implemented asynchronous processing for alerts and notifications using background job queues;
• Added authentication and security features, including JWT-based access control, session tracking, rate limiting, and audit logging;
• Contributed to frontend development, working with React and Next.js to support data visualization and platform usability

A web platform for archives that lets the public search, browse, and request archival materials online, while giving staff tools to manage orders and workflows. Users can view digital materials, explore archive hierarchies, and request visits, reproductions, or loans, with the backend handling user verification, order processing, and integration with existing archive systems via flexible data pipelines. A system for archives, museums, and galleries to catalog collections, track physical locations, manage loans, and attach related media. It supports hierarchical locations, custom metadata, audit logs, workflow automation, and advanced search with entity linking, enabling staff to efficiently manage complex collections.

• Developed clean and maintainable backend applications using Django;
• Designed and implemented modern, testable frontend features with Vue.js;
• Utilized Docker to containerize applications and improve deployment scalability;
• Worked with PostgreSQL to manage relational data efficiently;
• Integrated Apache Solr to enhance search functionality and performance;
• Resolved customer support tickets and addressed technical issues;
• Built ETL pipelines to manage and process customer data;
• Maintained Kubernetes clusters to ensure stable application operation.

A modern and performant frontend for an audio mastering platform using Next.js and React with TypeScript for type safety. Built a clean user interface with Material-UI and scoped styling, managed state with Zustand and React hooks, and implemented authentication with JWT-based session handling. Integrated interactive waveform previews, robust form handling with validation, payment processing, and internationalization support.

• Built the full frontend for an AI-powered audio mastering platform from scratch;
• Implemented user authentication, including Google OAuth login and secure session management;
• Integrated Stripe for subscription payments and premium feature access;
• Developed interactive audio waveform visualizations to allow users to preview tracks before and after mastering;
• Managed file upload workflows and display of live audio processing results;
• Ensured responsive and user-friendly interface, handling all aspects of UI/UX for track management and playback

This was an AI data annotation project focused on evaluating LLM-generated Excel solutions. The product involved a pipeline where annotators reviewed whether the model produced correct and prompt-aligned Excel formulas, and determined if a Python script was needed instead. When the task required it, Furkan wrote Python scripts that generated the expected Excel output programmatically. The goal was to improve the model's ability to handle spreadsheet-related tasks accurately, serving an AI training company refining its model through expert human evaluation.

• Reviewed LLM-generated Excel formulas for correctness and alignment with the given prompt.
• Wrote Python scripts using libraries like openpyxl and pandas to generate the expected spreadsheet output when formulas fell short or the task was too complex for native Excel functions.
• Annotated edge cases where the model misunderstood the prompt or chose the wrong approach entirely.

This was an AI data annotation project focused on evaluating and improving LLM-generated Python code. The product involved a structured pipeline where annotators assessed code outputs from competing models, identified the stronger solution, and refined outputs to meet higher standards of correctness, readability, and performance. It was built for an AI training company aiming to improve model quality through expert human feedback.

• Evaluated and compared Python code outputs from multiple LLMs, selecting the better-performing solution and annotating the reasoning behind each decision.
• Wrote optimized Python scripts for complex tasks, tested LLM outputs for correctness and performance, and provided detailed feedback to improve model training data.
• Handled edge cases and multi-step problems that less experienced annotators could not.

Аn early-stage deep learning platform for image classification and recognition, featuring a production-ready face recognition pipeline that leveraged ResNet50 and FaceNet for accurate identity verification in real-world conditions. The platform included foundational backend infrastructure and an initial user interface, as well as a distributed decision-making framework that fused perception data from multiple sources to improve situational awareness.

• Led end-to-end development of an edge AI platform, from architecture design to production deployment;
• Developed and maintained backend services using Node.js and Flask;
• Designed and implemented a Stream Engine for real-time processing of multiple video feeds;
• Built and optimized an Inference Engine for deploying deep learning models on edge devices;
• Created cross-platform client applications for Windows, Linux, and macOS;
• Collaborated with hardware teams to ensure compatibility across diverse edge devices