<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Uzay Sozen — Blog</title><description>Writing on AI engineering, LLM tooling and shipping software to production.</description><link>https://uzaysozen.dev/</link><item><title>Perfumind</title><link>https://uzaysozen.dev/projects/perfumind/</link><guid isPermaLink="true">https://uzaysozen.dev/projects/perfumind/</guid><description>An AI fragrance discovery platform I built solo, end to end: a Next.js web app, a FastAPI/LangChain backend and Supabase, with LLM-orchestrated research and recommendations across a catalogue of 4,000+ perfumes.</description><pubDate>Mon, 12 Oct 2026 00:00:00 GMT</pubDate><category>project</category><category>Next.js</category><category>FastAPI</category><category>LangChain</category><category>Supabase</category><category>pgvector</category><category>UMAP</category><category>Docker</category><category>Coolify</category><category>Polar</category><category>PostHog</category><category>Grafana</category></item><item><title>IMDb MCP Server</title><link>https://uzaysozen.dev/projects/imdb-mcp-server/</link><guid isPermaLink="true">https://uzaysozen.dev/projects/imdb-mcp-server/</guid><description>A Model Context Protocol server that gives AI agents structured movie and TV data through 23 tools, with an LRU cache that cuts upstream API calls by 50–75%.</description><pubDate>Mon, 12 Oct 2026 00:00:00 GMT</pubDate><category>project</category><category>Python</category><category>MCP</category><category>FastMCP</category><category>Caching</category><category>Docker</category><category>uv</category></item><item><title>Grape Detection with Mask R-CNN</title><link>https://uzaysozen.dev/projects/grape-detection/</link><guid isPermaLink="true">https://uzaysozen.dev/projects/grape-detection/</guid><description>My BEng dissertation at the University of Sheffield: a custom instance-segmentation model that finds grape bunches in vineyard images and video, reaching an 82% F1 score at IoU 0.50 with 0.3 s inference.</description><pubDate>Mon, 12 Oct 2026 00:00:00 GMT</pubDate><category>project</category><category>TensorFlow</category><category>Mask R-CNN</category><category>Computer Vision</category><category>Python</category><category>OpenCV</category><category>Flask</category></item><item><title>NetProVis</title><link>https://uzaysozen.dev/projects/netprovis/</link><guid isPermaLink="true">https://uzaysozen.dev/projects/netprovis/</guid><description>My MSc project: a web platform that deploys cloud-native network functions on Kubernetes and continuously retunes their autoscaling from ML forecasts, reducing cost by 5–10% on fluctuating workloads.</description><pubDate>Mon, 12 Oct 2026 00:00:00 GMT</pubDate><category>project</category><category>Kubernetes</category><category>GKE</category><category>Machine Learning</category><category>LSTM</category><category>TensorFlow</category><category>Python</category><category>FastAPI</category><category>React</category><category>Helm</category><category>GCP</category></item><item><title>Word to PDF Converter</title><link>https://uzaysozen.dev/projects/word-to-pdf-converter/</link><guid isPermaLink="true">https://uzaysozen.dev/projects/word-to-pdf-converter/</guid><description>A Word-to-PDF conversion service built as five microservices on Kubernetes, with RabbitMQ queues for asynchronous processing, MongoDB GridFS for file storage and JWT authentication. Designed to absorb up to 50,000 conversion requests.</description><pubDate>Mon, 12 Oct 2026 00:00:00 GMT</pubDate><category>project</category><category>Kubernetes</category><category>Docker</category><category>RabbitMQ</category><category>MongoDB</category><category>MySQL</category><category>Python</category><category>Flask</category><category>JWT</category></item></channel></rss>