Skip to main content

Use Cases

Semantic search goes beyond keyword matching it finds items that are conceptually similar to a query. With Ahnlich DB, embeddings generated by NLP models (e.g., sentence-transformers, OpenAI, BERT) can be stored and queried.

Example: A news website wants to allow readers to search for articles by meaning, not just keywords.

  • Query: “climate change effects on agriculture”

  • Stored vectors: embeddings of articles from the archive.

  • Ahnlich DB retrieves the most similar vectors using cosine similarity, filtering out irrelevant categories (e.g., topic != sports).

GETSIMN 5 WITH [0.42, -0.13, 0.76, ...] USING cosinesimilarity 
IN news_store WHERE (topic != sports)

This returns the top 5 most semantically similar articles, even if the exact keywords differ.

2. Recommendation Engines​

Recommendation systems depend on finding similar users or items. By storing user embeddings and item embeddings, Ahnlich DB can power real-time recommendations.

Example: An e-commerce platform wants to recommend products to a user based on their browsing history.

  • User embedding: [0.24, 0.15, 0.93, ...] (from their profile + past behavior).

  • Product embeddings: vectors representing catalog items.

  • Query: Find the 10 most similar products to the user’s embedding.

get_sim_n(
    store="product_store",
    search_input=[0.24, 0.15, 0.93, ...],
    closest_n=10,
    algorithm=CosineSimilarity,
    condition=Predicate::NotEquals {
        key="status",
        value="out_of_stock",
    }
)

This returns personalized recommendations, excluding unavailable items.

3. Clustering & Exploration​

Clustering groups together vectors that are close in high-dimensional space. Ahnlich DB makes this possible by serving as a backbone for unsupervised learning workflows.

Example: A research lab stores embeddings of thousands of genomic sequences. To discover patterns:

  • Query vectors representing sequences with certain traits.

  • Run similarity searches to group them into clusters.

  • Use metadata filters (species = human) to narrow analysis.

This allows scientists to explore relationships between data points without explicit labels.

4. Filtering with Context​

Ahnlich DB allows developers to combine vector similarity search with predicate-based filtering to return results that are both semantically relevant and contextually constrained.

Example: A video platform wants to recommend “similar” videos but only those uploaded in the last 12 months.

find top-8 similar videos
GETSIMN 8 WITH [0.89, -0.33, 0.55, ...] USING euclideandistance IN video_store;

filter results by upload_date
GETPRED (upload_date > "2024-09-01") IN video_store;

In this workflow, GETSIMN identifies the most similar vectors, while GETPRED applies metadata conditions. By chaining them together, applications ensure that retrieved results are not only “similar” but also satisfy real-world constraints such as time windows, categories, or user-specific rules.

5. Cross-Domain Retrieval​

Since embeddings can represent any modality—text, images, audio—Ahnlich DB supports multi-modal retrieval.

Example: A fashion company wants customers to upload a photo of an outfit and find similar products in their catalog.

  • Image converted into an embedding.

  • Query against the vector database (fashion_store).

  • Metadata filter ensures only items in stock are suggested.

GETSIMN 3 WITH [0.31, 0.88, 0.64, ...] USING cosinesimilarity 
IN fashion_store WHERE (availability = "in_stock")

This enables image-to-product search, a powerful e-commerce feature.