Original architectures and systems developed by SKD RATWANI Technology & Software Solutions Pvt Ltd. Documented here as a public record of innovation — with GitHub commit history as technical verification.
May 2026
Putaway Intelligence — Quantitative Warehouse Layout Research
A configurable warehouse simulation framework for evidence-based slotting and layout decisions. Compares I-shape, U-shape, and (roadmap) L-shape / modular layouts across multiple putaway strategies — random_stow, abc_slotting, distance_minimized, and (Phase 4) reinforcement learning.
Key innovation: Generic configurable profiles + vectorized order generation enable SMEs to validate layout decisions in hours, not months — at zero consulting cost. Same SKUs/orders run across layouts make the trade-offs explicit and reproducible.
First case study (May 2026): I-Shape vs U-Shape comparison across 200 SKUs, 34,476 order lines, 800 receipts, 310 locations. Result: 6.7× spread between worst combination (I-shape + distance-minimized) and best combination (U-shape + ABC slotting).
WMS Simulation
Reproducible Research
SME-Accessible
Read the case study →
2025 – 2026
Generic Warehouse ML Simulation Framework
A three-layer reinforcement-learning pipeline for warehouse putaway research. Layer 1: Isaac Sim perception → CLIP embeddings → Neo4j knowledge graph. Layer 2a: Gymnasium PPO policy with curiosity-driven exploration. Layer 2b: Unity ML-Agents environment with procedural warehouse simulation and GAIL imitation learning from expert demonstrations.
Key innovation: Industrial-grade RL pipeline accessible to SME operations — bridges classical heuristic strategies (ABC slotting) with learned policies, enabling fair quantitative comparison on shared synthetic data. Same Putaway Intelligence dataset becomes the evaluation contract.
Status: Framework built and integrated. Training in progress (currently at step 36,978/target 80,000). NVIDIA Inception application pending for production training credits.
GAIL Bridge
Isaac Sim
Unity ML-Agents
PPO+Curiosity
April 2026
Wiki Intelligence Engine
A self-improving knowledge system where every quality-verified AI response automatically enriches a shared knowledge base. Platform gets faster and more accurate with every interaction — without model retraining.
Key innovation: Three-tier quality gating (0.75 / 0.95 thresholds) with human oversight at the critical middle band. AI handles extremes. Humans verify edge cases.
AI Architecture
Knowledge Systems
EdTech
2025 – 2026
SEAL — Self-Evolving Adaptive Learning
An observer-based learning pipeline that evaluates every AI response against domain quality criteria, assigns scope (personal / pending / global), and accumulates verified knowledge independently of the underlying model.
Key innovation: Separation of knowledge accumulation from model training. Platform intelligence grows without retraining.
Reinforcement Learning
Quality Gating
Self-Improvement
2025 – 2026
Domain Factor Intelligence (ADAPT)
Multi-vertical AI architecture serving WMS, IBM iSeries/AS400, Civil Construction, and Apparel — each with fine-tuned adapters, domain-specific data models, and entity-aware response calibration.
Key innovation: Same question returns different resources for an individual learner vs a university vs an organisation. Relevance is contextual, not universal.
LoRA Fine-tuning
Entity Awareness
Multi-vertical AI
April 2026
Partnership-First Content Ethics
A principled approach to AI content sourcing — external knowledge is indexed only after explicit partnership agreement. Every content source shows users our relationship status and how it was verified.
Key innovation: GTM pipeline integrated into the platform — discovery → outreach → agreement → indexing as a single tracked workflow with full user transparency.
Content Ethics
GTM Pipeline
Transparency
2025–2026
SkillCode — Software Engineering powered by AI
An AI-native code intelligence tool that lets development teams chat with their codebase in plain English — understanding structure, dependencies, and risk without reading source files manually.
Supported codebases: IBM iSeries/AS400 (RPGLE, CLLE, CLP, DDS), Node.js / Express.js, and internal enterprise platform codebases.
iSeries-specific capabilities: Job Intelligence using STRSRVJOB — trace active jobs, analyse execution paths, understand service program relationships, and surface runtime behaviour that static code review misses.
Core capabilities across all codebases: Dependency mapping, modification risk scoring, file relationship analysis, multi-level content indexing (files, functions, classes, routes, procedures), Job Trace UI for visual execution analysis, and natural language chat over any indexed codebase. Supports multiple codebases simultaneously with separate intelligence per project.
IBM iSeries
Node.js
Job Intelligence
Job Trace
Codebase Chat
Dependency Analysis
Multi-codebase
Job Trace UI
April 2026
Research Studio
An admin-facing research pipeline that imports curated domain resources from any source (manual, Claude web search, Perplexity), runs AI audit on each resource, and routes to human review before publishing.
Key innovation: Multi-source research with quality scoring and human oversight before any content reaches learners.
Research Pipeline
AI Audit
Human Oversight