1. Discovery on the collections and the systems behind them

We began with a short, focused discovery phase to understand two things:

• The collections and how those items are described and managed
• The systems where records lived, including multiple WorldCat-based repositories and additional collection platforms with different schemas and maturity levels.

This helped us define a realistic integration strategy that accounted for the structure and operations of the disparate collection management platforms while delivering a unified experience to the public.

2. A practical federation strategy: unify discovery in one index

Federated search is typically built in two ways:

• Search-time merging (query each system live, then combine results), or
• Index-time unification (normalize key fields and index everything centrally).

For AHC, we chose index-time unification. The reason was simple: While query-time mappings would have been faster to set up, we were concerned about performance and providing the most relevant results. We also wanted to take full advantage of all the features offered by the search engine we chose.

Our strategic plan was to map each system to a shared set of common metadata fields (title, description, date, etc.) which was sufficient to support powerful search and filtering.

3. Choosing a search engine built for speed and long-term maintainability

Once the central index approach was defined, we evaluated search engines and selected Algolia. We were attracted by Algolia’s performance, ease of use, and the platform’s tooling made it efficient to iterate on the model and relevance while keeping the solution maintainable over time.

4. A robust ingestion pipeline, built to evolve

On the technical side, the most complex work wasn’t “search” itself—it was building reliable extraction routines and managing change over time across heterogeneous systems. This included answering practical questions like how to detect updates and deletions from each source.

Key elements of the build included:

• Nightly ETL batch jobs that refresh the index on a reliable cadence and can be rerun if a job fails, ensuring missed updates are picked up without manual backtracking.
• Full reindex capability that allows AHC to reload the database from scratch when needed, providing a dependable path for recovery and reprocessing as systems evolve.
• No-downtime updates that run behind the scenes so routine nightly refreshes do not affect the front-end experience.
• Source-level refresh isolation so a complete refresh of an individual system may briefly affect that source for a few minutes, while overall search remains available and results repopulate as items are re-ingested.
• Public-ready indexing controls that ensure only records fit for public consumption are included (for example, excluding items not yet ready for public view).
• Media caching for performance that stores images and PDFs so content loads quickly and consistently within the search experience.
• Monitoring and notifications that alert administrators to extraction or ingestion issues quickly, supporting operational confidence over time.
• A cloud-based foundation on AWS using serverless components to improve reliability, scalability, and long-term maintainability.

5. Designing for both exploration and precision

AHC’s audiences range from casual visitors to historians and researchers with highly specific needs. In our research, we focused on real-world behaviors like visitors exploring the museum’s collections, students of all ages working on papers, and location-based research; and we grounded the experience in the kinds of requests users bring to AHC staff.

To create an inviting entry point, we helped shape a landing experience built around popular research paths including historical events and historical themes. From there, the search experience supports users with facets and filtering, allowing users to begin with a simple keyword and then narrow intelligently. Designing this “simple-to-start, powerful-to-refine” experience (including the facets, mixed record types, and results grid) was one of the most meaningful parts of the work.