Human-built. Data-backed. AI where it helps.
HawkShift is a shopping platform for finding, comparing, and timing real product buys — especially furniture, kitchen, coffee gear, and electronics. It is not a marketplace: we do not take your payment or ship boxes. We curate a live catalog, run our own data pipelines, and send you to retailers via affiliate links when you are ready to purchase.
Three things sit under every feature: a human who designed and ships the product; data from real products and price scans in our database; and AI used deliberately — for language, planning, and some editorial generation — never as a substitute for inventory that does not exist.
Built by a human
HawkShift is a solo project. The architecture, catalog pipelines, Concierge agent, price-intelligence jobs, account system, ops tools, and UI were designed and implemented by one developer — also the founder of an AI automation practice. There is no large content team rewriting listicles weekly, and no black-box “AI shopping startup” layer you cannot inspect in practice.
It started as a personal system to stop overpaying and guessing. It became a full site when the same problems kept showing up: keyword search that lies, prices that move without warning, and “AI” tools that invent products instead of retrieving them.
The catalog — 17,539 active products · 2 retailers
The heart of HawkShift is a PostgreSQL catalog of 17,539 active products across 5 categories. Products are imported and enriched from retailer feeds and APIs. One retail path is the primary checkout conversion; other retailers may appear in the data for pricing and coverage where we have them.
Each product can carry images, titles, categories and subcategories, ratings when the source provides them, price rows per store, affiliate URLs, and history. We do not invent SKUs. If it is not in the database, the Concierge cannot honestly recommend it as available on HawkShift.
Data engines — search and ranking
Everyday shop search is deterministic. A shared search warehouse parses your query for keywords, price bounds, and category or subcategory signals, then filters active, in-stock products and ranks them with transparent rules: subcategory match, category match, full-text relevance, rating, and a preference for the primary affiliate retail path when it helps conversion — without deleting other product rows forever.
On top of that structure, we use style and attribute labeling for furniture and related goods (so “bohemian” or “minimalist” can mean something more than a lucky keyword hit), and vector embeddings as a fallback for open-ended phrasing. The goal is fast, grounded retrieval — not a generative model writing a shopping list from thin air.
Price intelligence — history and signals
Prices are scanned on a regular cadence across retailers we track. Each update can write to store price rows and price history. From that history we compute deal-oriented signals used on product pages and in the Concierge: how today’s price sits relative to recent and longer windows, and verdict-style guidance such as stronger buy vs wait when the model has enough history.
These are statistical summaries of data we collected — not guarantees that a price will drop tomorrow. Always confirm the live price on the retailer before you buy. Our live ticker and alerts (1 active watches in the system) exist to surface movement as scans complete, not to replace checkout.
Cross-store matching — 412 UPC links
Where the same product can be identified by UPC across retailers, we store a match and compare prices. When our primary affiliate retail path is cheaper by a meaningful margin, we may show that on shop cards and product pages. When another retailer is cheaper, we stay quiet in the UI until we have a proper affiliate path — we do not push unpaid checkout traffic. We never delete product pages to “merge” catalogs; old URLs should not 404 for matching alone.
AI Concierge — agent + tools + interface
The Concierge is a visual field plus chat: product nodes, compare overlays, and full furnish plans on a floor-plan layout. Under the hood it is a hybrid system. Language models help with conversation and planning, but actions are structured — search the warehouse, show nodes, open a pair compare, emit a multi-room furnish plan, clear the field. The interface applies those actions instead of free-drawing the UI from model prose.
Session memory keeps budget, style, and prior intent so “something cheaper” or “darker” can mean something coherent. For home furnishing, a dedicated planner allocates a total budget across rooms with weighted shares, then pulls real catalog products per subcategory and style band. You can save a plan to your account and reopen it later.
Room visualize uses generative image models to sketch a room from plan context. That path is credit-limited and best-effort: pretty pictures are optional; the plan and prices remain the product.
Comparison — live tool and 25,204 indexable pages
The live compare experience puts products side by side: specs, prices, ratings, and deal context. Separately, we maintain 25,204 indexable head-to-head pages so people can land from search on a specific matchup. Editorial on those pages may be AI-assisted; prices still come from our database as it updates.
Accounts, alerts, and email
Sign-in is magic-link email (optional Google when configured). Your account can hold saved furnish plans and price alerts tied to your login. Alerts notify you when a watched product hits a target after our scans. We also run a newsletter of real deals from our catalog, filtered into sensible discount bands — not fake 99% off noise. You can unsubscribe anytime.
How we make money
HawkShift participates in retail affiliate programs. If you buy through our links, we may earn a commission at no extra cost to you. That is disclosed site-wide. We design features to help you decide well; commissions are how the lights stay on while the catalog and tools stay free to use.
What this is not
Not a store. Not a guarantee of the lowest price on earth. Not an LLM that invents inventory. Not a claim that every number is perfect the second a retailer changes a tag. It is a human-built layer of data, software, and carefully used AI to make high-consideration shopping less wasteful.
Questions or feedback? Reach us at hello@hawkshift.com or visit our contact page. See also our affiliate disclosure and privacy policy.