Muhammed Murat Yanaşoğlu

SiteScribe consolidates scattered construction evidence (site logs, RFIs, plan revisions, and images) into a single platform and turns it into traceable, audit-ready Change Order workflows. I built a multi-tenant system that centralizes field logs, RFIs, plan revisions, and photos; detects signals to generate events; and produces evidence-backed Change Order drafts. It includes role-based access control (RBAC), a notification system, comment/approval flows, PDF and ZIP export, and optional Mistral AI integration for end-to-end enterprise workflows.

I developed a Django web application that extracts Latin, Ancient Greek, and Fraktur text from stone inscriptions and historical printed sources using Tesseract OCR with script-specific image preprocessing. Google Gemini integration produces English translation, summaries, and historical context. The project supports export for research use in TXT, JSON, Word, and PDF formats and exposes a REST API.

I built an AI-guided trading automation system that analyzes crypto and stock markets—including U.S. SEC Form 4 insider filings—to detect buy/sell signals. The platform also generates structured table- and chart-based reports that summarize trading patterns and highlight notable activity for faster decisions.

RegLens monitors regulatory sources for changes, analyzes them, and maps updates to internal controls, tasks, and audit evidence so compliance runs from one panel. I delivered an end-to-end RegTech platform with AI-assisted normalization, matching, and evidence suggestions; task and audit-pack management; role-based access control (RBAC); security logs; rate limiting; and PWA support.

A desktop learning app focused on bilingual vocabulary through real-time object recognition. It uses ROI-controlled Focus Mode and a two-stage detector–classifier stack (YOLOv8n + MobileNetV2), delivers 30+ FPS on CPU only, runs fully offline, and preserves privacy.

Artworks can be analyzed in depth with a chosen AI model or compared across multiple models. You can retrieve analysis outputs, model metrics, and accuracy scores, and download them in HTML and JSON. Automatic blog posts can be generated with a selected tone and length. Camera input is supported.

A web marketplace that connects people seeking services with providers across Northern Cyprus and Turkey in a secure, fast, and transparent way. I implemented a scalable backend with NextAuth (credentials + Google OAuth), role-based access (customer, provider, admin), real-time messaging, notifications, profile and bidding flows, and an admin panel (user/service approval, location management, statistics, audit log). The UI uses React with Tailwind CSS and Bootstrap; reporting uses Chart.js. Optional Mistral AI powers an in-platform assistant.

Two separate portals for students and academics. Students view availability and submit appointment requests; faculty manage requests, approve or reject them, and communicate via messaging.

A desktop application for monitoring screen time, controlling usage, and applying security locks with advanced monitoring. It tracks usage patterns, enforces time limits, and provides detailed analytics for digital wellness.

An AI-powered, real-time voice assistant for hotel guests. It uses the Web Speech API for multilingual speech recognition and text-to-speech, and Google Gemini plus Mistral AI for dynamic conversations. An admin dashboard gives managers full access to update content. Built with Next.js, React, TypeScript, Prisma ORM, and MySQL.

A full-lifecycle AI platform that automates the process from data collection and evidence management through drafting, revision, and final petition packages. Multi-LLM orchestration (OpenAI, Claude, Gemini) handles generation and quality control, with multilingual voice assistant and chatbot support.

Public examples of AI-assisted personal portfolio sites.