Tugay
Akdemir
Tugay Akdemir is a 23-year-old embedded systems and AI developer based in İzmir, Turkey, building production software at the intersection of hardware, computer vision, and energy infrastructure. At Werover, he develops drone inspection tooling — GPT-4o Vision damage classification, DJI .MRK GPS telemetry parsing, and automated blade-position mapping integrated into panel.werover.ai. He engineers IoT products including MeshLink, a Heltec ESP32-S3 + SX1262 LoRa mesh platform for GSM-dead wind farm sites with solar power and IP67 field deployment. On the full-stack side, he ships multi-tenant SaaS platforms SunFusion (renewable O&M, BESS modules, Supabase RLS) and the GES Energy Dashboard (EPİAŞ API, real-time sell-opportunity analytics for 100 MWp+ solar portfolios). Graduating from Ege University Electronics and Communications Technology in June 2026; open to embedded, IoT, and full-stack roles across Turkey and Europe.
End-to-end vision and geospatial pipeline integrated into Werover's production drone inspection workflow. Ingests and parses DJI .MRK GPS telemetry files, extracting per-frame latitude, longitude, and altitude to align inspection photos with physical blade geometry. A Z-altitude pass-detection algorithm segments each flight into three distinct inspection passes — Pressure Side (PS), Leading Edge (LE), and Suction Side (SS) — before interpolating span-wise coverage and computing precise 0–100% blade position for every photo frame. GPT-4o Vision API performs automatic damage detection with CAT 1–5 severity classification, structured CSV export, and downstream reporting. Validated across 5 turbines and 18+ MRK files in field conditions.
Dual-mode product architecture spanning rugged field hardware and an operator-facing dashboard, built for GSM-dead wind farm and industrial SCADA environments. Hardware stack: Heltec ESP32-S3 host MCU, Semtech SX1262 LoRa radio, custom LoRa mesh networking protocol, solar panel input with LiPo battery management IC, and IP67 weatherproof enclosure for pole-mount, plug-and-play deployment. Firmware engineered for TEİAŞ SCADA protocol compliance to interface with Turkish grid telemetry requirements. Targets B2B wind energy operators lacking reliable cellular backhaul. Awarded 3rd place at Ege University Proje Pazarı.
Multi-tenant renewable-energy SaaS platform on Next.js 14 and TypeScript with strict tenant isolation. Authentication layer uses JWT session tokens with bcryptjs-hashed credentials; authorization enforced via Supabase Row Level Security policies on every tenant-scoped table. SuperAdmin panel provisions organizations, module entitlements, and user roles. Six energy vertical modules — Solar O&M, Solar Denetim, BESS O&M, BESS Denetim, Solar Keşif, Trafo Bakım — gated behind role-based access control (RBAC). EPİAŞ market data ingestion, 8,760-hour annual hourly dataset analytics, and operational workflow automation. Production deployment at sunfusion.vercel.app.
Real-time solar production versus consumption analytics dashboard built for Kontek Enerji, managing 100 MWp+ utility-scale portfolios. EPİAŞ Transparency Platform API integration with 60-second polling interval for live day-ahead and intraday market price feeds. Ingests and processes full 8,760-row annual hourly datasets for yield benchmarking, curtailment analysis, and loss attribution across sites. Automated sell-opportunity detection engine evaluates net surplus (netFazla) against spot price thresholds — triggering actionable export alerts when surplus generation exceeds zero and market price rises above 3.50 TL/kWh. Next.js, TypeScript, and Recharts power interactive time-series visualization.
- (—)Architected GPT-4o Vision API pipeline for wind turbine blade damage detection: multimodal image ingestion, structured prompt engineering, and CAT 1–5 severity classification with confidence scoring per defect region.
- (—)Built interactive labeling and review tool for inspection analysts — batch upload, annotation overlays, and exportable CSV datasets feeding model iteration and client deliverables.
- (—)Developed Python–PowerPoint automated report pipeline: parses inspection CSV/JSON outputs, maps damage coordinates to blade position metadata, and generates client-ready slide decks with per-turbine summaries.
- (—)Implemented MRK GPS file parser correlating DJI telemetry timestamps with photo EXIF data; Z-altitude pass-detection algorithm segments PS/LE/SS passes and computes 0–100% blade position across 18+ MRK files on 5 turbines.
- (—)Integrated tooling into Werover production platform (panel.werover.ai); demonstrated live end-to-end workflow at industry trade show with field-captured blade imagery.
- (—)Deployed ESP32-S3 based acoustic monitoring hardware in turbine nacelle environments — firmware bring-up, sensor interfacing, and field validation alongside vision inspection stack.
- (—)Developed ESP32-based Modbus TCP to RF (UART) smart gateway bridging Ethernet-attached SCADA masters to license-free sub-GHz radio clients in industrial plants.
- (—)Implemented FreeRTOS multi-task architecture with dedicated Modbus parser, RF transmit/receive, and watchdog tasks; thread-safe FIFO queue decouples bursty TCP register reads from constrained RF airtime.
- (—)Handled Modbus Exception 06 (Slave Device Busy) responses with exponential backoff and request coalescing; Cache Mode serves stale register snapshots during RF link recovery to prevent upstream timeouts.
- (—)Optimized Data Packing: aggregates adjacent holding registers into single RF frames, reducing round-trip latency ~40% versus per-register polling on lossy links.
- (—)Deployed 1 master + 4 client PLC topology — Mitsubishi FX3U slaves on RS-485, ESP32 gateway as protocol translator; validated register maps against HMI SCADA screens in factory acceptance testing.