Nineteen Projects. One Way of Thinking.
Small, focused systems I built to understand commerce problems from first principles — ranking, recommendation, root-cause analysis, grounding, performance and security. Each has an interactive demo with hard-coded, synthetic data (the SAP Commerce Developer Toolkit links to its live app instead); the code is on GitHub.
Flagship Work
The projects that best show how I think about commerce systems.
Revenue Detective
Answers “why did revenue fall?” with ranked, evidenced causes instead of dashboard archaeology.
Recommendation Engine
Hybrid recommender where every suggestion carries the reason it was made.
Product Search Engine
Structured filters plus a hand-written BM25 ranker over a 124-product catalog.
Research Agent
A plan → search → evaluate → synthesise pipeline that cites evidence or admits it has too little.
API Performance Lab
Reproducible before/after benchmarks for indexing, caching, pagination and parallelism.
Commerce API Security Lab
A defensive commerce API: rate limiting, JWT, RBAC, ownership checks and audit logging, mapped to OWASP.
Strong Supporting Work
Focused engines, each with a clear algorithm and a clear business use.
Churn Prediction API
A from-scratch logistic regression returning churn probability, risk tier and top contributing factors.
Transfer Network Analysis
Graph analysis of a synthetic transfer market: PageRank, corridors, fee outliers, flurries and boomerangs.
Dynamic Pricing Simulator
Rule-based pricing from demand, inventory, competitor and seasonality — bounded by margin and move-size guardrails.
Fulfillment Optimizer
Chooses the cheapest SLA-aware fulfilment plan across a warehouse network, splitting shipments when needed.
Product Discovery Engine
Turns a natural-language shopping query into structured filters, with a trace of where each value came from.
SAP Commerce Developer Toolkit
A live, browser-only toolkit with two SAP Commerce generators: an OCC Controller Generator & Tester and an ImpEx Generator.
SAP Commerce MCP
A read-only MCP server that lets Claude or another MCP client inspect a local SAP Commerce instance through HAC, in natural language.
Engineering Experiments
A shared pattern: deterministic guardrails for catalog and content — the layer that makes LLM output safe to ship.
Catalog Quality Auditor
Seven deterministic checks that score catalog data quality: duplicates, gaps, mismatches and suspicious claims.
FAQ Generator
Generates FAQs strictly from supplied product data and flags any answer the data does not support.
SEO Description Generator
A staged content pipeline that generates SEO copy and checks it against the product facts.
Review Summarizer
Extracts sentiment, themes and recurring complaints, with every finding tied to review IDs.
SEO Score Calculator
A fully deterministic six-part SEO score with no LLM.
Match Intelligence
Turns raw match rows into team form and plain-English insights.
Reading this honestly: most of these are deliberately deterministic — rules, statistics and classical algorithms. Only the churn model is machine learning, and no demo calls an LLM. The SAP Commerce MCP demo is a simulation with sample data, not a live connection. The “LLM-ready” projects show the grounding and validation layers an LLM would sit behind.