Fourteen systems, each measured on a real run - latency, accuracy, test counts, sweep results. Every entry opens into the problem, the approach, the numbers and the screenshots.
- Swarm · multi-agent orchestration with machine acceptance 11/11 checks passed 4 turns 21 с reference run 21 merged branches
- Engineer Companion · offline RAG over medical-equipment documentation 21,214 facts indexed 96/100 benchmark accuracy 130 ms p50 latency · CPU 768/775 tests passing
- GeoConverter · numerical geodetic calibration <1 mm round-trip error 20+ local coordinate systems Commercial release, government clients
- VETKA_DWG · survey sketches → CAD drawings 4h→15m manual drafting time 3 surveying companies live 444 GUGK blocks in the library
- PenaPivo / BeerLog · serverless Telegram bot + Mini App ≥84% branch coverage 3 UI languages Workers edge runtime
- VTuber Rigger · AI auto-rigging: model or screenshots → VRM 1.0 53 bone VRM skeleton 17 expressions 2 input paths
- Neuro Property Trade · deterministic board-game engine + custom UI 0 hardcoded tiles Godot 4 deterministic engine 100% replayable
- Razgon Quiz · live quiz-show platform (hall, remote, builder) 386 precache entries 98 mp3 cues 30 с question fuse
- Hermes Viz · telemetry console for an agent farm 2 292 graph nodes 839 edges 21 agents 9 models
- Codebot · Telegram bridge to a coding-agent runtime 63 classes 10,9k LOC Java 21 no dependencies
- Storybeam · AI video pipeline (active fork) 56 commits +14 955 lines added 108 files
- Trading research station · systematic strategy validation 1.44 PF, 4h 2 526 trades 82% CPCV consistency 33 strategies swept
- Streaming stateful prediction · a research problem 4–5× stronger signal 1.38× graph headroom 500k rows scanned
- Static websites · hand-built, no libraries 2 sites 6 pages 390 px mobile width
- Based
- Timezone
- Availability
- English
- Focus