INTERACTIVE PROOFS OF CONCEPT

Applied Architecture, Enterprise Workflows & Industrial Systems

Each project is an interactive proof of concept built to test a specific idea under real conditions: deterministic routing before model inference, governed AI workflows with a full audit trail, and the operational constraints of enterprise and industrial systems. Every demo runs in the browser, so you can inspect the inputs, the decisions, and the evidence behind them.

ENTERPRISE AI & GOVERNANCE POC

Verified Demo

Roseberry Consulting Foundry

Governed Enterprise AI for Procurement Triage & Contract Auditing

Problem & Scope

Enterprise operations teams field a steady stream of IT hardware requests and multi-page vendor invoices, and most of them follow rules that never need a language model. This proof of concept applies a "Deterministic SQL First, AI by Exception" architecture: every request and invoice is evaluated locally against deterministic SQLite rules at $0.00 and 0 tokens first. Only ambiguous scope notes, catalog gaps, repeat-spend clustering, cybersecurity holds, and semantic contract clause disputes escalate to structured AI reasoning.

Architecture & Key Capabilities

01
Hardware Triage: Evaluates ServiceNow RITM requests against local item master catalogs, routing clean matches directly to Oracle Cloud ERP requisitions while escalating off-catalog specs and security-sensitive gear to AI triage and Cyber Gate Holds.
02
Contract & Invoice Auditor: Audits vendor invoices against 6-page IT MSP and Construction MSAs, combining Stage 1 deterministic SQL rate math with Stage 2 AI semantic clause interpretation, interactive tap-to-scroll contract citations, and automated AP short-pay clawback vouchers.
03
Hybrid Architecture & MCP Trace: Exposes the full trust boundary across every run, including exact prompts, Model Context Protocol (MCP) tool calls, SQL vs. AI execution splits, and schema-validated outputs.

Deterministic SQL + AI
MCP Tool Routing
ServiceNow & Oracle ERP
MSA Clause Auditing
FastAPI & SQLite

Launch Foundry Verified Demo

INDUSTRIAL AI & DETERMINISTIC ROUTING POC

Interactive Benchmark

OT Context Engine

Deterministic Context Routing for Local LLMs

Problem & Scope

Explores how locally hosted language models handle dense industrial, pipeline, and cybersecurity standards without context-window dilution or vector-chunking loss. Instead of dumping entire manuals into a prompt or relying on fuzzy similarity search, a deterministic hierarchical router isolates the exact regulatory sections and Markdown tables required before model execution.

Architecture & Key Capabilities

01
Side-by-Side Benchmark: Compares Path A (raw document stuffing) against Path B (deterministic section routing) across token consumption, time-to-first-token, total execution latency, and citation accuracy on local GPU hardware.
02
Regulatory & Industrial Corpus: Indexed across 8 authoritative documents spanning NIST SP 800-82r3, CISA OT cybersecurity advisories, PHMSA 49 CFR Parts 192 & 195 pipeline regulations, and ISA/IEC 62443 frameworks.
03
Document Preparation & Section Routing: Demonstrates structure-preserving PDF-to-Markdown conversion, hierarchical section pruning, and verifiable page-level grounding.

Local LLM Inference
Hierarchical Routing
NIST / PHMSA / IEC 62443
Structure-Preserving Markdown
Zero-RAG Determinism

Launch OT Context Engine