SECTOR 07 — RESEARCH

THE LAB_

From-scratch AI research, run on my own hardware. Small generative models taken apart and put back together — each experiment ships an illustrated explainer, and most come with a hands-on demo you can poke at in the browser.

capyos terminal — sector 07
EXP_01 :: ppe-detection
OBJECT DETECTION

Site Safety Detection

A fine-tuned YOLO detector that picks out people and their safety equipment on construction sites — helmets, hi-vis vests, gloves — and flags when someone is working without them.

object detectionYOLOfine-tuningcomputer vision
DEMOIN PROGRESS31 July 2026
EXP_02 :: vector-world
GENERATIVE GEOGRAPHY

Vector World

An illustrated atlas of a generative model that writes geography as a program — a 33.7M-parameter transformer expanding a map one 500 m tile at a time, with continuity across every shared edge enforced by contract rather than hoped for. Grow a world yourself inside the page.

transformersDSLvector tilesOpenStreetMapprocedural
EXPLAINERINTERACTIVE27 July 2026
EXP_03 :: outpaint
CANVAS EXPANSION

Outpaint

A 92M-param diffusion U-Net trained from scratch to extend images past their borders — masked DDIM sampling, candidate batches, stride experiments, and an interactive replay of a real 16-tile expansion run.

diffusionv-predictionoutpaintingfrom scratch
WRITEUPINTERACTIVE26 July 2026
EXP_04 :: terramind-flood
EARTH OBSERVATION

TerraMind Flood Mapping

Flood segmentation from Sentinel-1 SAR with the TerraMind foundation model, fine-tuned on Sen1Floods11. A staged weak-label pipeline, real IoU numbers — and an honest look at why 4,160 free labels made things worse out of distribution.

geospatialSARsegmentationTerraMindTerraTorch
WRITEUPIN PROGRESS25 July 2026
EXP_05 :: diffusion
IMAGE GENERATION

Diffusion From Scratch

A DDPM trained on CIFAR-10 in ~250 lines of plain PyTorch — 18.7M-parameter U-Net, linear noise schedule, DDIM sampler, EMA weights — plus an ONNX export verified against PyTorch for in-browser WebGPU inference.

DDPMU-NetPyTorchDDIMONNX
DEMOIN PROGRESS22 July 2026
EXP_06 :: llm-maker
LANGUAGE MODELS

LLM Maker

Training a language model from scratch on a single GPU — tokenizer, architecture, multi-hour resumable training runs, and what the loss curve actually teaches you.

LLMtrainingPyTorchfrom scratch
WRITEUPIN PROGRESS10 July 2026

// 6 experiments logged — more are compiling. build notes live in the wiki