contextsearch by contextforce
contextsearch · live on 3,433 products

Search that gets it.

Keyword search matches words.Semantic search matches meaning.Both miss the intent: the problem, the budget, the brand they mean.contextsearch reads the intent and ranks by what buyers actually weigh. More searches end in a sale.

Try the demoSee it compared

Same catalog. Two answers.

    Keyword search

    BM25: stop words, stemming, field boosts

    contextsearch

    Same keyword search + intent layer

    Real results from the local index, graded blind by an AI judge from query and title alone. Times are server time (median of 15 runs, repeat query, intent plan cached, local machine). A never-seen phrasing adds one LLM planning call, about 0.2 to 0.6 s locally (median 0.5 s over 10 queries). Both lanes land together.

    Keyword

    Matches words.

    "overheating" Xbox cables and games.

    Semantic

    Matches meaning.

    "under $15" can't read a price.

    contextsearch

    Matches the need.

    "overheating" cooling stands.

    Shoppers say what they need. Catalogs say what it is.

    We connect the two.

    Learn the catalog once. Read every query live.

    1

    Sort it.

    Every product gets a type.

    2

    Read it.

    Facts, and what it is good for.

    3

    Match it.

    Each search ranked by the job.

    One approach. Every category.

    Sorts products correctly

    vertical cooling stand → console cooling fan

    Reads the details

    left 4 dead 2 → shooter · mature

    Knows what each product does

    active cooling · vertical stand

    Understands problems

    "my xbox keeps overheating"

    Understands plans and occasions

    "take my controller on a trip"

    Respects limits

    "controller under $15"

    Knows your nicknames

    "TB headset" → turtle beach

    Finds more like this

    "games like halo"

    Shows its reasoning

    why: advanced cooling · adjustable fan speed

    The numbers.

    Top-5 hit rate (P@5) · 40 queries
    keyword0.72
    context0.93
    Ranking quality (NDCG@10) · same queries
    keyword0.59
    context0.83
    Index · 3,433 products
    ≈ $0.63

    First full index.

    Per new search
    ≈ $0.0002

    New phrasing, about 0.5 s. Repeats, milliseconds.

    Internal benchmark on a title-only Amazon Xbox 360 catalog. Queries mix plain keywords, problem-phrased and use-phrased searches; relevance graded blind by an AI judge and repeated from fresh caches. The keyword baseline is a standard setup: English stop words removed, plurals folded, title weighted above brand and category, prefix on the last word and small typo tolerance. Results on your catalog and query logs will differ.

    Next.

    Shopping lists"Skiing, as a beginner" becomes jacket, boots, layers.
    Second opinionAn optional check on the top results.
    Shopper modifiersBeginner, budget, for kids.
    More departmentsAbout 670 product types drafted.

    See it on your catalog.

    Try the demo