henrique mandrick

I've built computer vision, LLM, web, and mobile projects while studying Computer Science at FIAP, including Verge Studio, a mobile-video pipeline that measures height in real-world metric scale without markers, LiDAR, or depth sensors.

For Verge Studio, I provisioned cloud GPU infrastructure and connected multi-frame depth inference to an automated measurement harness, processing 94 frames in 31.5 seconds. I also implemented browser-based semantic segmentation with SlimSAM-77 through Transformers.js and WebGPU.

I created Mapillary Grass Height from the ground up, collecting images, producing semantic segmentation masks, annotating 1,987 road-edge crops, and evaluating an ordinal regression model. When the subjective labels and lack of real-world scale limited the approach, I proposed and developed Verge Studio.

My BotSolar work won first place at FIAP's NEXT/GOODWE competition. I rewrote its energy-automation agent around LangChain and Groq, built Tuya tools with secure logging and unit tests, and contributed to GoodWe and FastAPI integrations; before technology, I spent five years in administrative and financial operations.