Interactive Technical Lab
Experiments Across WebGL, NLU, GIS & Construction Math
Working tooling built to eliminate repetitive manual work across architectural visualization, site quantity takeoffs, geospatial traffic modeling, and bilingual voice interfaces.
Semantic Hierarchy glTF Extractor
Blender 5.2 Python pipeline grouping architectural collections into mutually exclusive semantic GLTF nodes (FullWalls3m vs SectionWalls1m, SlabAndFinishes, StairsAndBalconies, OverheadSlab, Furnishings) so Three.js toggles cutaways and exploded views without duplicate downloads.
- Compresses geometry via Draco level 6 + WebP 85 textures (reducing Cluster 01–03 assemblies to 3.47–4.46 MB)
- Centers bounding boxes on X/Z with Z=0 floor datum while exporting JSON dimension manifests
Deterministic English + Hinglish NLU Engine
Zero-hallucination natural language parser built for LoadCheck that translates mixed Hindi-English phrases ("aur ek bedsheet daal do", "remove 2 shirts", "kitna weight") into typed Dexie IndexedDB operations guarded by plausibility checks.
- Pure TypeScript stages: IntentDetector -> NumberResolver -> ReferenceResolver -> OperationPlanner -> SemanticSafetyGate
- Interactive sandbox runs directly in your browser on the Laundry Manager case study
SiteCore Pure-TS Quantity Engine
Versioned construction quantity formulas (`concrete.v1`, `masonry.v1`, `plaster.v1`, `steel.v1`) separating net geometric BOQ quantities from site wastage ordering factors.
- Guards against invalid opening deductions exceeding gross wall surfaces
- Calculates cement bags (1,440 kg/m³), steel unit mass (π/4 · d² · ρ), and brick counts live in-browser
QGIS-MCP & Gangapol Traffic Graph
Connects PyQGIS to coding agents over FastMCP (118 tools) alongside a strict-no-guessing OpenStreetMap corridor traffic survey script for Chokdi Gangapol, Jaipur.
- Filters <15m OSM stubs and classifies segments into LOW / MODERATE / HIGH V/C only when observed widths & counts exist
- Unobserved corridors stay explicitly marked UNASSESSED
Weed-Detection-Ai-model
Starter public GitHub repository (`akshat00dev/Weed-Detection-Ai-model`) linked to Veltrionex Smart Creations Pvt Ltd’s agricultural crop-vs-weed detection robotics initiative (May 2025 – July 2026).
- Public repository: akshat00dev/Weed-Detection-Ai-model (starter README)
- Paired with `dhobimate-skills` & `indodes` on GitHub
Interactive Live Engines
Test both deterministic engines directly in your browser without leaving the portfolio.
Formula: Vg = L × W × T ; Vn = Vg − D ; Vo = Vn × (1 + waste% / 100)
Audited Concrete BOQ Breakdown
Net BOQ Volume (Vn)
3.800 m³
Order (+5% waste)
3.990 m³
Engine runs purely in TypeScript (`sitecore_v1` IndexedDB) with immutable snapshots.
Workflow · Calculate → Explain → Save to Project → Price → Record Work
All 7 Built-In SiteCore v0.1 Calculators & Golden Verification Checks
Slabs/beams/columns minus deductions · Ex: 5×4×0.125m = 2.500 m³
Bricks & wet mortar minus openings · Ex: 5×3×0.23m = 2.967 m³ (1,558 units)
1/2-face plaster + cement/sand · Ex: 2 faces @ 12mm = 25.8 m² (0.3096 m³)
N-coat paint area + 10% allowance · Ex: 12.9 m² × 2 ÷ 10 + 10% = 2.838 L
Whole tiles & boxes minus cutouts · Ex: 5×4m (600mm) = 62 tiles, 16 boxes
Circular rebar π/4×d²×7850×L×N · Ex: Ø12mm × 10m × 5 = 44.391 kg
Exact metric-imperial helper · Ex: 12 ft 2 in ↔ 3.7084 m
INR totals separated from physical units · OVERRUN (>100%) alerts · JSON/CSV
Estimation Aid Disclaimer: Quantities, mortar splits, and rates are geometric planning estimates only — not structural design, mix design, Bar Bending Schedules (BBS), BIS/byelaw compliance, or certified contractor billing. Always verify quantities against approved contract drawings.
Pipeline Output Trace
Ready · Deterministic pipeline mirrors loadcheck IntentDetector.ts