Skip to main content

Searching in Datasets

Once datasets are ingested, Amplifi allows you to efficiently search through your files. This feature helps you extract relevant insights quickly by querying the ingested datasets.

Accessing the Search Page​

  1. Navigate to the Search Tab
    Go to your Workspace and select the Search tab from the sidebar.

  2. Select Datasets
    Use the Dataset dropdown to choose one or more datasets you want to search through.

  3. Enter Your Query
    Type your question or search prompt in the query field.

  4. Search Method
    Amplifi supports multiple vector similarity metrics to rank results:

    • Cosine Distance: Measures directional alignment between embeddings; great default for semantic relevance.
    • L2 Distance (Euclidean): Measures straight-line distance; useful when embedding magnitudes matter.
    • Lp Distance: Generalized distance metric; set p to tune sensitivity (e.g., p=1 is L1, p=2 is L2).

    By default, Cosine Distance is used unless configured otherwise.

  5. Run the Search
    Click the Search button to view results ranked by relevance.

Understanding the Results​

Each search result includes:

  • A snippet of matched content from the dataset
  • A Search Score, indicating how closely it matches your query (higher is more relevant)

Viewing Performance Metrics​

At the top-right, you’ll see a checkbox: Show metrics per dataset

✅ When enabled:​

You’ll see separate metrics for each dataset:

  • Precision: Measures how relevant the results from that dataset are to your query.
  • NDCG (Normalized Discounted Cumulative Gain): Reflects the quality of result ranking — higher means the top results are more relevant.
  • Latency: Time taken to return results for the search (in seconds).

This helps you compare how well each dataset is performing.

⛔ When disabled:​

You’ll see a single aggregated view showing overall:

  • Precision
  • NDCG
  • Latency across all selected datasets.

Summary​

  • You can search across multiple datasets using semantic similarity.
  • Results are ranked using vector similarity metrics (Cosine, L2, or Lp).
  • Performance metrics help assess dataset quality and search effectiveness.

Use this interface to find accurate insights, helping you quickly find relevant data for further analysis.