JITI Dashboard — Manufacturing Inventory Intelligence
Replacing manual spreadsheet workflows with automated 12-month projections, CNY blocking detection, and AI-generated ordering recommendations for international and domestic manufacturing inventory control.
The Challenge
The client distributes automotive parts from overseas manufacturers to dealers across North America. Their inventory planning process was almost entirely manual — a senior buyer would receive spreadsheets from multiple sources (warehouse inventory, QuickBooks, purchase order trackers, ocean/air shipment logs, production schedules), open them side by side, and spend hours calculating whether each product line had enough stock to survive the next 12 months.
The complexity was not just math — it was context. Chinese New Year shuts down factories for weeks, and if you have not placed orders 85+ days in advance (45 days production, 35 days ocean transit, 5 days customs), you are out of stock for months. Different product classes (A/B/C by volume) have different target inventory levels. Container quantities matter because you cannot order 1,500 units when a container holds 2,000 — you either fill the container or pay for dead space.
The recommendations themselves had to be formatted in a specific way — “OE REC 4000 PCS (2 CONTAINERS) SHIP 2026-04-15” — because that is the language suppliers understand. Without a system, the buyer was the system. If that person went on vacation, inventory decisions stopped. The business could not scale its planning capability beyond one person's cognitive bandwidth.
Our Approach
We built the JITI Dashboard as a clean-architecture .NET 8 application that replaces the spreadsheet workflow with an automated pipeline: import data from 11 Excel sheet formats, calculate Months of Inventory (MOI) across stock classes, project 12 months forward with CNY blocking detection, and generate AI-powered ordering recommendations in the exact format suppliers expect.
The architecture uses four layers — Domain, Application, Infrastructure, and Web — with strict separation enforced by project references. The Domain layer defines entities for Products, Inventory Positions, Projections, Purchase Orders, and Shipments. The Infrastructure layer contains the heavy lifting: supply chain services handle projections and container calculations, rules services manage MOI calculation with CNY-adjusted targets, and the Excel import service maps 11 different spreadsheet formats into normalized domain entities.
The AI integration was designed as a recommendation layer, not a replacement for domain logic. The analysis service builds structured prompts that include the 12-month forward projection, current MOI status, risk classification, and container constraints — then asks Claude (with Gemini and OpenAI as fallback providers) to generate actionable ordering recommendations. The output is the exact REC CAPS format that buyers use to communicate with suppliers.
Multi-tenancy was a late addition that proved architecturally clean: the system supports two product lines with tenant-level isolation via Entity Framework global query filters and separate UI pages for the secondary product line, which requires dual MOI calculations (firm-only vs combined with forecast).
The Solution
Excel Import
11 sheet formats normalized into unified domain model
Calculation Engine
MOI, stock class rules, CNY blocking, container logic
AI Analysis
Claude + Gemini + OpenAI generate REC CAPS recommendations
Dashboard
12-month pipeline, risk colors, alerts, export to suppliers
Excel Import Engine
11 configured sheet formats — DS Inventory, QB Inventory, QB Purchase Orders, production schedules, ocean/air trackers, and more. Column mappings via configurable JSON. Every entity links back to its import log for full lineage tracking.
MOI Calculation Engine
Months of Inventory calculated per product line with status classification: Critical (<1) through Excess (>12). Stock class targets — Class A (3-6 MOI), Class B (4-8), Class C (6-12) — with automatic CNY adjustments adding 1.5 months during the pre-CNY window.
12-Month Projection Engine
Forward inventory balance: opening + incoming supply - projected demand = closing. CNY blocking flags months where factory closures prevent new orders. Must-order-by dates calculated from lead time chains (production + ocean + customs = 85+ days).
Container Logistics Calculator
Converts order quantities to container counts using per-product unit capacity. Outputs in the REC CAPS format suppliers expect: “OE REC 4000 PCS (2 CONTAINERS) SHIP 2026-04-15” — ready for direct copy-paste into supplier communications.
AI Recommendation Service
Constructs structured prompts with current MOI, 12-month projections, CNY windows, lead times, and container constraints. Multi-provider architecture: Claude primary, Gemini and OpenAI as automatic fallbacks. Output formatted for direct supplier communication.
Integrated Help System
10 searchable help topics, 5 FAQs, 2 quick-reference cards, and 35+ field-level tooltips via Bootstrap popovers. Enables self-service onboarding without training sessions — critical for a domain this specialized.
Results
0 errors, 0 warnings
Build status
4-layer Clean Architecture
Architecture
11 Excel sheet formats
Import formats
12-month forward view
Projection horizon
3 with auto-fallback
AI providers
5 of 5 — complete
Phases delivered
Inventory planning time reduced from hours of manual spreadsheet work to minutes of automated analysis. CNY ordering risk eliminated — the system calculates must-order-by dates automatically based on lead time chains.
REC CAPS recommendations generated in supplier-ready format — no reformatting needed. Multi-tenant architecture supports both product lines from a single deployment. The comprehensive help system enables self-service onboarding without training sessions.
The pipeline dashboard provides a visual early-warning system — red months are immediately visible, allowing buyers to focus attention where it matters rather than scanning spreadsheets for anomalies. The AI recommendation layer turns raw projection data into actionable supplier communication, bridging the gap between analytics and execution.
“The hardest part of inventory intelligence isn't the math — it's encoding the domain knowledge that lives in one buyer's head. CNY timing, container fill rates, stock class thresholds, lead time chains — this system doesn't replace the expert, it captures their expertise so the business doesn't depend on a single person being at their desk.”
Frequently Asked Questions
Have Domain Knowledge Trapped in Spreadsheets?
We build systems that capture expert workflows and turn them into scalable, automated intelligence.