ISCV | Isocurve Research Lab (@iscvlab) is a multidisciplinary spatial data, GIS cartography, research synthesis, and computational design laboratory created and maintained by Napat (Tan) Phasundhiae (@napattan) in Bangkok, Thailand. We engineer deterministic software tools, spatial analytics pipelines, and parametric workflows for the built environment.
| Pillar | Domain Focus | Key Capabilities & Stack |
|---|---|---|
| 01 // Research Synthesis & Strategy | Evidence-Based Research & Feasibility | Systematic Literature Reviews • Policy & Regulatory Synthesis • Environmental & Sustainability Frameworks (LEED/WELL) • AI Research Workflows |
| 02 // Spatial Data & GIS Mapping | Multi-Scale Cartography & Urban Analytics | QGIS Desktop & PyQGIS • Space Syntax Network & Visibility Analysis (SSE) • GeoAI Canopy Segmentation (SAM 3 + DeepForest 2.1) • Spatial SQLite & GeoPackage Pipelines |
| 03 // Data Analysis & Visualization | Statistical Modeling & Visual Information Design | Exploratory Data Analysis (EDA) • Executive Dashboards • High-Precision Vector Plates • Analytical Diagrams & Publication Decks |
| 04 // Computational & Parametric Design | Algorithmic CAD & Environmental Simulation | McNeel Rhinoceros 3D & Grasshopper • RhinoCommon (C# / Python) • Multi-Threaded Microclimate & Solar Solvers • Design-Tool Automation |
| Repository & Pillar | Status | Core Capability & Empirical Impact |
|---|---|---|
CanopyGIS (GitHub • Org Mirror)02 // Spatial Data & GeoAI |
LIVE (v1.0.0 • Live Demo) | Hybrid Tri-Engine urban tree canopy & treetop centroid extraction (SAM 3 + Weecology DeepForest 2.1.0 + Compact Geodesic Watershed + 3-Pass RAG) across On Nut (229.4 Acres (928,176 sqm), 2,987 crowns), Nong Khaem (150.2 Acres (607,888 sqm), 2,110 crowns), and Ramintra Sports Park (23.4 Acres (94,544 sqm), 1,432 crowns) with < 30 ms real-time physical filter sliders (foliage_prob, relief_3d_score, artificial_cyan_score) and -100% (< 4.0 m²) micro-shard elimination (1,568 → 0). |
sse-space-syntax (Org Mirror)02 // Spatial Data • 04 // Design Tech |
LIVE (v1.1.1 • QGIS Portal) | Dual-engine Native Rust Standalone Workstation (sse-desktop), CLI (sse-cli), 9-Tool MCP Server (sse-mcp) & Qt6-Ready PyQGIS Plugin executing exact UCL depthmapX v0.8.0 Tulip-1024 directed-segment parity (R² = 1.0000, 54.4×–57.8× faster Angular Segment Analysis; 16.69 s on Bangkok 166,514 segments in < 16 MB CSR RAM; 34.9×–144.3× faster Axial; 35.8× faster VGA & Analytical Isovists; 495.5× faster Gibson Visual Agents). |
xlaya01 // Research & Semantic AI |
LIVE (v2.4.0 local • Release v2.3.0) | Standalone 5-profile System 1 → System 1.5 semantic decision, regime-invariant 8D AST/code-graph sieve (6,237 B LF sieve_genomes.json), Wallacei v2.7 K-Fold OOF NSGA-II MOO (22/22 pytest, 12/12 --self-test), & conformal triage engine (laya 0.4.31+ & vegaml 0.16.1+), pairing --profile nano (0 PyTorch/NumPy, 8.50 MB clean nano RSS, 3.74–3.99 ms warm / 274.25 ms/repo uncontended v2.4 cold [255.33 ms/repo v2.3], 84.05% disjoint val 31B R@8 / 70.83% multi-hop / 0.0091 ECE, 65.12% E4B R@3 / 79.07% 31B R@8 on blind Kaggle 129 R1, 78.72% R@8 on SWE-bench_Verified, 81.37% Val / 71.69% MCQ nvidia/When2Call conformal tool gate, 86.21% PLE 29) with 6-tool resident stdio MCP warm inference (~30 ms), 859.1 q/s GPU batch (8.9× mixed EN/HI), 0.45/0.55 ensemble jury, --thorough 4-facet rubric + file_summaries, --vega-views 1..3 debiasing, xlaya_fit persistent TTTHead (90.0%), local multimodal, and 0.00% XERJ #1108 wrong-and-confident fires. |
xboost01 // Research & Workflows |
LIVE (v3.6.0 local • Release v3.5.0) | Composable 15-skill /x* ecosystem and 10 deterministic Python CLI & MCP engines (22,237 LOC) compatible with any AI coding agent — cloud, IDE, CLI, or local LLM (Claude, Grok, Gemini, Cursor, Codex, Ollama, OpenHands, Cline, etc.), featuring the 4-Tier Sharded Compound Learning Playbooks (L1 Hub + 4 L2 Shards) (/xlearn, 24 PAT-*/OPT-* across 21 families incl. OPT-MOO-01, 13 FALS-01..13, 7 GOV-01..07), /xmouse Background Ghost Mouse, 100% red-team defect recall across 25 seeded defects in 8.14 ms, -97.06% briefing context noise reduction, and 5-profile /xlaya v2.4.0 (8.50 MB clean zero-torch --profile nano, 79.07% Blind Kaggle 129 R1 & 78.72% SWE-bench_Verified 31B R@8, 71.69% When2Call MCQ, ~30 ms warm 6-tool MCP / 859 q/s GPU batch / 0.45/0.55 ensemble / --vega-views 1..3 / 90.0% vega-U routing) passing XERJ #1108 with 0.00% wrong-and-confident fires. |
xocr01 // Research & Document AI |
LIVE (v2.4.0 local • Release v2.1.0) | High-speed C++ PyMuPDF vector/table extractor (19-97 ms/page, 15.0×-3,326× faster hybrid routing), Two-Stage LaTeX booktabs & scanned HTML <table> → RFC-4180 CSV surgery, Thai PUA/Sara-Am normalizer, and 5-profile local SCB 10X Typhoon-OCR v1.5 (Qwen3-VL) & baidu/Unlimited-OCR-2 engine. |
Upstream PR Contributions01–04 // Open-Source R&D |
LIVE (24 PRs + 1 Hold • 10 MERGED • 13 Orgs) | Empirical open-source bottlenecks & bugfixes powered by XBoost (Full Benchmark Playbook): • Geospatial & Vision: Weecology/DeepForest #1449 (MERGED), #1447 & #1464 ( #1440) (74×–421× Shapely 2.0 vectorization) · opengeos/geoai #960 (MERGED) · gboeing/osmnx #1412 (MERGED, closed #1410) & #1405 (40.2× speedup) · martibosch/pylandstats #77 (4.0× centroids) · opengeos/segment-geospatial #569• Climate, ESG & Urban Physics: CLIMADA #1350 (MERGED) & #1349 ( 69.9× apply_box / 11.7× CSR impact_at_reg & #1331 World Bank ISO fallback) · ladybug-tools/uwg #292 (1,735.6× speedup & #291 custom ref_bem_vector climate zones) · Deltares/hydromt #1566 (27.5× speedup & #1563 non-EPSG:4326 basin CRS) · os-climate/physrisk #651 (17.5× Chronic Heat) · natcap/invest #2813, #2829 (Hold; 23.1× UCM & 18.1× UFRM; #2839 Closed) · UMEP-dev #885, #149, #6 (3 Staged: 173.0× UTCI, 799.8× load_walls)• Rust & Agent Infrastructure: xerj-org/xerj #2922–#2927 (6/6 MERGED) ( 6 PRs: skill root auto-indexing, Thai PDF PUA, MSVC fixes, 0.00% false-confident gate on #1108) · scb-10x/typhoon-ocr #16 (OPEN, #15 Closed, -35.9% vision tokens) · mufeedvh/code2prompt #348 |
mcp-bridges04 // Computational Design |
Supporting Utility (8 Bridges • Selected for XERJ Corpus Hub) | Supporting Model Context Protocol (MCP) automation bridges connecting any cloud, IDE, CLI, or local AI agent to AutoCAD, QGIS, Rhinoceros 3D, Grasshopper, Adobe Illustrator, and Figma (8 skills incl. Ghost Mouse Companion) with 1-prompt universal agent install and Parallel-Brains Single-Mouse /xultra swarm safety (144 resolved xerj autoindex graph edges). |
qgis-axonometric-transformer02 // Spatial Data & GIS |
Supporting Utility (v1.2.2 • QGIS Portal #6219) | Supporting Qt6-verified (QGIS 4 Ready) plugin for 1-click 3D isometric, dimetric, and oblique axonometric map diagrams (30× faster than manual CAD) with multi-format transparent clipboard export (PDF /SMask, SVG, PNG, CF_HDROP) to Adobe Illustrator & Figma. |
bangkok.iscvlab.com02 // Spatial Data • 03 // Analytics |
WORKING PROTOTYPE (Secondary / In Polish • 8 Dashboards A–H) | Interactive city-scale Bangkok Macro Spatial Intelligence Suite (A–H) backed by tiles.iscvlab.com Cloudflare R2 PMTiles vector tiles covering transit accessibility, environmental comfort, land use, and urban ecology. |
| Initiative | Pillar | Status | Scope |
|---|---|---|---|
environmental-simulation (Iscv.EnvSim) |
04 // Computational Design |
PREPARED | Standalone GPU & CPU environmental physics engines (D3Q19 LBM Wind CFD, ISO 9613-2:1996 8-Band Outdoor Acoustics, Tregenza/Reinhart Perez Solar & SVF Raycaster) validated against AIJ/CEDVAL, ISO/TR 17534-3, and CIE 171:2006 benchmarks. |
| Timberpin Field Sensing | 04 // Design Tech • IoT |
PREPARED | Handheld environmental telemetry & acoustic tree tomography ground-truthing hardware (ESP32 + Bosch BME688). |
- Creator & Open-Source Maintainer: Napat (Tan) Phasundhiae (
@napattan) • Personal LinkedIn (in/napatphas) - Official Channels (
@iscvlab): LinkedIn (company/iscvlab) • Instagram (@iscvlab) • Facebook (@iscvlab) - Inquiries & Collaborations: tanfasantia@gmail.com
