Sixteen deep-dive guides and interactive tools on the ML, experimentation, SQL, and analytics that actually move metrics in e-commerce — written by Vikas Sharma, Data Scientist at Noon. No vendor fluff, no listicles.
Each guide stands alone, but together they form a curriculum: from the map of the field, to the daily toolkit, to the specialist plays.
From SELECT * WHERE title LIKE '%query%' to hybrid semantic retrieval — we build an e-commerce search engine in 8 levels, with working code at every step, a live demo you can type into, and an honest verdict on when to build and when to buy. By the end, you'll have the complete picture.
A complete theory guide for building a Multi-Tenant Knowledge Base SaaS product. From first principles to production architecture, with 60+ interview questions.
Not another "8 use cases of AI in retail" listicle. This is a map of the machine learning that actually runs e-commerce today — which algorithm wins each problem, why, and what to learn — written from inside the industry.
Every e-commerce business question is a variation of about eight SQL patterns. Master these — funnels, cohorts, RFM, repeat rate, aging, sessionization — and you can answer almost anything leadership asks, same day. Written in BigQuery SQL, portable everywhere.
Most A/B tests in e-commerce don't fail statistically — they fail organizationally: underpowered designs, peeking, metric fishing, and results nobody acts on. This is the full playbook for running experiments that produce decisions instead of arguments.
Industry estimates say 50–60% of trade promotions lose money. Almost nobody knows which of theirs do, because almost nobody measures incrementality. This is the full playbook: margin math, holdout design, uplift targeting, and the segment strategies that follow.
Every retailer has a mountain of stock that stopped selling, and a P&L quietly bleeding storage cost, trapped cash, and eventual write-downs. This is the analytics playbook for seeing the problem clearly, pricing the exit correctly, and — the hard part — acting early enough.
Everyone built a shopping assistant; the real money is quieter. Catalog enrichment, review mining, support triage, semantic search, competitive intelligence — where LLMs actually pay for themselves in e-commerce, with the prompts, the cost math, and the evaluation discipline the listicles skip.
It's 7am, GMV is down 5% vs last week, and leadership wants to know why before the 10am call. This is the systematic answer: the metrics tree, the rate-vs-mix math, the traps, and a step-by-step runbook that turns panic into a diagnosis.
Searchers are ~24% of visitors and ~44% of revenue, converting 2–3x better than browsers — yet most companies run search as a black box someone configured in 2021. This is the operating playbook: the KPI stack, query mining, merchandising rules, and the quality process.
A 1% improvement in price realization is worth more than a 1% traffic increase, at zero marketing cost. This is the analytics that finds it: KVIs and price image, competitive indexing done honestly, a practical elasticity recipe with all its traps, and the governance that makes price changes stick.
Fashion e-commerce runs 25–40% return rates; COD markets add 20–40% RTO on top. Between reverse logistics, damaged stock, and refund costs, returns quietly consume more margin than most promo budgets — and almost nobody measures them properly. Here's the full analytics playbook.
You paid for the traffic, won the search, earned the add-to-cart — and then a declined card or a COD refusal took the revenue anyway. Checkout and payments are where 60–70% of intent dies, and almost nobody staffs the analytics. Here's the playbook.
From zero to a working multi-tool agent. Environment setup, tool creation, agent orchestration, callbacks, session management, and multi-agent systems — with real code you can run today.
From my research assistant's rate-limit error to knowledge graphs — what memory actually means for an LLM agent, why each technique exists, when to reach for it, and the working code. Eight levels, one real bug as the running example, no hand-waving.
A tool is how an agent stops being a chatbot. This is the practitioner's dictionary: 28 tools across 8 shelves, the masterclass on building custom tools models actually use well, MCP, 10 e-commerce use cases with working code, and the security rules that keep it all safe. Pick, copy, ship.
Calculators for the two decisions that come up most often.
Find the discount depth where a promo stops paying for itself.
How many users you need before the result means anything.