Analytics foundation → AI-assisted prototype. A supplier-risk decision workflow built around trusted metrics, evidence, and human judgment.
1. Problem & user need
Supply Chain Health Monitor used SQL and Tableau to surface quality, inventory, lead-time, logistics, and profitability signals across 100 SKUs and five suppliers. It answered “What is happening?” but an analyst still had to connect the signals, explain their significance, and decide what to investigate.
The Copilot explores that interpretation gap. The primary user is a procurement or supplier analyst; a procurement manager is a secondary reviewer. The objective is decision support, not autonomous purchasing.
100 SKUs
5 suppliers
Static portfolio dataset
AI-assisted development
2. My role & responsibilities
I framed the user problem, defined the MVP and product requirements, prioritized capabilities, designed the workflow and trusted supplier context, and developed the prototype through AI-assisted development. The portfolio work combines supply-chain domain knowledge, product decisions, and technical implementation.
The accompanying product artifacts include a PRD, roadmap, prioritization and trade-offs, UX definitions, a Figma prototype, technical architecture, and a proposed evaluation framework. These are product-design artifacts; their presence does not imply that every proposed feature has been implemented.
3. Discovery & requirements
Job to be done: When reviewing supplier performance, help an analyst understand which supplier needs attention, why, and what evidence to investigate next, without stitching the story together manually from several charts.
The research basis is my operational experience, workflow analysis, and prototype exploration. Formal primary research with procurement users and production usability validation have not yet been completed.
The MVP focuses on selecting one supplier, reviewing calculated metrics, generating an evidence-backed AI brief, and retaining human control. Supplier comparison, decision-history controls, and feedback capture are broader product proposals rather than current prototype capabilities.
4. Planning, architecture & key decisions
I prioritized the supplier risk brief over contract retrieval, a tool-using investigation agent, or a full procurement platform. It addresses a specific interpretation problem with a smaller implementation scope and fewer consequential actions.
Architecture: cleaned data → deterministic analytics → structured supplier context → Netlify function → OpenAI API → structured risk brief → human review.
Risk tiers are calculated from supplier-level average defect rates. The model receives the calculated tier and relevant quality, delivery, inventory, and exposure metrics. Negative lead-time deviation means faster than the portfolio benchmark, not a delivery failure.
The application uses a defined JSON output schema and enforces the deterministic risk tier and human-review requirement. Prompts require evidence and forbid invented contracts, purchase orders, costs, or unsupported SLA claims. These controls guide output; they do not prove that every generated statement is correct.
5. Implementation & prototype walkthrough
This project is proof that I can take domain knowledge from supply chain, apply AI product thinking, define system boundaries, and use AI-assisted development to turn the concept into a working prototype.
I originally built a Supply Chain Health Monitor using SQL and Tableau to surface supplier quality, inventory, and lead-time signals. The dashboard answered “What is happening?” but an analyst still had to manually answer “Which supplier needs my attention, why, and what should I investigate next?”
I redesigned that workflow into an AI-assisted Supplier Intelligence Copilot. Supplier risk is calculated deterministically. AI is used for interpretation, evidence-grounded explanation, healthy-signal identification, and recommended investigation.
Data provides facts. Deterministic logic establishes risk. AI explains and synthesizes. Humans decide.
The original Supply Chain Health Monitor combines SQL business rules and Tableau views for inventory health, supplier performance, revenue and profitability, logistics efficiency, and defect risk. Its rule-based classifications remain the analytical foundation for supplier and SKU-level exception analysis.
Current capabilities
Supplier selection and source-derived operational metrics.
Deterministic supplier risk classification, kept separate from original SKU-level classifications.
Focused supplier context and a structured AI risk brief.
Risk drivers with evidence, healthy signals, advisory investigation steps, and data limitations.
Human review before consequential sourcing decisions; no procurement actions are executed.
6. Results, limitations & future work
The result is a working portfolio prototype with a clear separation between calculations, AI interpretation, and human decisions. It is not a production procurement system. No measured reduction in analyst time, production adoption, or business impact is claimed.
The static dataset covers 100 SKUs and five suppliers. Risk thresholds are simplified for demonstration and would require business validation before operational use.
The evaluation framework is a proposal covering evidence grounding, signal interpretation, recommendation quality, missing-data handling, and structured-output compliance. It does not represent executed benchmarking or production evaluation results.
Future roadmap: approved tool calling, document retrieval/RAG, supplier comparison, enterprise authentication, ERP connectivity, approval workflows, and monitoring. These are not implemented integrations.